Currently showcasing 12 feeds. More coming soon!
MIT researchers have developed a security protocol that uses the quantum properties of light to ensure that data transmitted to and from a cloud server remains secure during deep learning computations. The protocol encodes data into laser light used in fiber optic communications systems, making it impossible for attackers to copy or intercept the information. The technique maintains a 96 percent accuracy rate while providing robust security measures.
1. Researchers at Carnegie Mellon University have developed Neural Motion Planning, an artificial intelligence network that allows robots to navigate unfamiliar environments without obstacles. The system uses data-driven simulations to train the robots to perform reactive and fast motion planning, allowing them to adapt and be more versatile.
2. In simulations, the robots encountered various household environments and random objects, such as shelves, dishwashers, and vases. The robots were able to avoid obstacles like lamps, plants, bookcases, and cabinet doors, demonstrating their ability to complete tasks in diverse household environments.
3. Neural Motion Planning provides a stepping stone in large-scale learning for robotics, especially in complex tasks like motion planning. This approach allows robots to learn and generalize from simulated environments to the real world, enabling them to navigate unknown settings successfully.
Boston-based company Neurable has developed a pair of smart headphones called the MW75 Neuro that use electroencephalography (EEG) and artificial intelligence to track the wearer's focus levels by reading their brain waves. The headphones capture 80 to 90% of the signals that traditional EEG technology can and can also play music. The device aims to improve mental wellness and prevent burnout, tracking focus levels and providing insights on how to improve work routines.
The headphones work best when the user is stationary, and users can earn "focus points" for every minute spent in high or medium focus. Neurable sets a goal of 100 focus points each day. The device can also detect when the user's focus is declining and can suggest breaks to avoid burnout. The device's brainwave data is converted into focus information and anonymized before being stored in a secure cloud database.
Cloudflare is launching a suite of tools that allow websites to monitor and selectively block AI data-scraping bots. These tools will be available to all Cloudflare customers, including the 33 million using its free services. The suite includes a real-time bot monitoring dashboard, expanded bot-blocking services, and the ability to pick and choose which bots to block or allow.
Cloudflare's bot-blocking measures will not be easily ignored by bad actors, as they compare it to having a physical wall patrolled by armed guards. The company has created processes to spot even the most carefully concealed AI crawlers.
Cloudflare is also planning to launch a marketplace for customers to negotiate scraping terms of use with AI companies, providing a way for content creators to receive value in return for their content. The company aims to ensure that humans get paid for their work and plans to facilitate licensing agreements and permissions arrangements between AI companies, publishers, and websites.
A committee of experts from top U.S. medical centers and research institutes is using NVIDIA-powered federated learning to train AI models for tumor segmentation in cancer detection.
Federated learning allows organizations to collaborate on AI model development without sharing sensitive data, addressing privacy and data management constraints.
The team is using NVIDIA FLARE and NVIDIA MONAI to optimize the training process and improve annotation accuracy, and plans to publish their methodology and pretrained model for future use.
Xavier Niel, a prominent figure in the French AI industry, has become a board member of TikTok's owner, ByteDance, as the company faces legal challenges in the US.
Niel, known for his disruptive approach, believes that Europe needs to invest in homegrown AI to compete with Asia and the US.
He has invested €200 million in French AI, launched a nonprofit research lab, and aims to develop AI infrastructure in France through his cloud provider, Scaleway.
A group of sex industry professionals and advocates is calling for inclusion in shaping AI regulations, stating that current discussions risk excluding their perspectives and overregulating their industry.
The group argues that policymakers need their insight to regulate in a way that protects fundamental rights and fosters a more sex-positive online environment, rather than risking censorship and misunderstandings.
The European Commission encourages adult industry representatives to participate in public consultations on AI regulations, including upcoming discussions on "unacceptable risks or prohibitions."
Researchers from the Chinese Academy of Sciences have developed a deep learning-based autofocus method for grayscale images.
The method uses a comprehensive dataset of grayscale image sequences and two focusing strategies to dynamically select regions of interest within the frame.
The deep learning-based autofocus method achieved fast and accurate focusing, demonstrating the potential of AI in enhancing traditional imaging technologies.
OpenAI has released a new series of AI models, called OpenAI o1-Preview, that are designed to improve reasoning and problem-solving capabilities.
These models have been trained to refine their thinking processes, try different methods, and recognize mistakes before providing an answer.
In tests, the models performed comparably to Ph.D. students in tasks related to physics, chemistry, biology, mathematics, and coding.
Apple's new iPhone 16 and 16 Pro models, with advanced AI capabilities, are expected to drive a boom in sales and start an iPhone "supercycle," according to some analysts. The AI features, such as improved photo editing, custom emoji generation, and a more natural Siri voice assistant, are seen as a major draw for customers. However, other analysts are skeptical of the supercycle prediction, noting that the AI features may not be compelling enough to drive large numbers of early upgrades.
This article provides a tutorial on creating a pronunciation assessment app.
The app's goal is to help users improve their pronunciation by allowing them to enter a word, record their voice, and receive a score based on the API's analysis.
Knowledge of JavaScript and Vue.js 3 is recommended to follow the tutorial.
AI researchers at Tsinghua University and Zhipu AI have developed a large language model (LLM) called LongWriter that can generate text output of up to 10,000 words.
Existing LLMs are typically limited to generating short outputs of around 2,000 words because they are trained on short documents. LongWriter overcomes this limitation by training on longer documents.
The researchers have made the code for LongWriter open-source, allowing others to build upon their work. They also posted a video demonstrating LongWriter generating a 10,000-word tourist guide.
Roeland Decorte, inspired by his experience in codebreaking, developed a smartphone app that uses AI to listen for signs of disease hidden in a person's pulse.
Decorte's company, Decorte Future Industries, is at the forefront of an audio-powered revolution in healthcare, using algorithms to interpret the body's faint signals and diagnose various conditions, including heart problems and stomach cancer.
By using just a microphone, Decorte's technology can provide accurate readings at home, eliminating the need for multiple apps and hardware solutions for different conditions.
Researchers have ranked AI models based on their risk level, revealing a wide range of behaviors and rule-breaking tendencies among these models.
Regulations may need to be tightened to address the legal, ethical, and regulatory compliance issues associated with AI models.
Government rules and guidelines on AI are found to be less comprehensive than companies' policies, suggesting room for improvement and the need for stricter regulations.
1. MIT researchers have developed SigLLM, a framework that uses large language models (LLMs) to detect anomalies in time-series data without the need for training or fine-tuning. LLMs have the potential to be more efficient and cost-effective than deep-learning models for anomaly detection tasks.
2. The researchers found that LLMs can convert time-series data into text-based inputs that the models can process. They developed two approaches, Prompter and Detector, that use LLMs to locate anomalous values and predict future values, respectively.
3. While LLMs did not outperform state-of-the-art deep learning models, they showed promise in anomaly detection and could be used to flag potential problems in equipment such as wind turbines or satellites before they occur. Future work will focus on improving LLM performance and understanding their performance in anomaly detection tasks.
Scientists from MIT's Computer Science and Artificial Intelligence Laboratory (CSAIL) have developed a method to train robots in simulated environments created using "digital twins" of real-world spaces. Users can scan a physical environment using a smartphone app and make necessary adjustments to create a digital replica. Robots can then train in the simulated environment, which accelerates the learning process compared to training in the real world.
The approach eliminates the need for extensive reward engineering and can create strong policies for various tasks, improving performance even in environments with disturbances and distractions. The researchers tested the system's performance in controlled lab settings and real-world environments and found that it outperformed traditional imitation-learning methods, especially in situations with visual distractions or physical disruptions.
RialTo, the robot training system developed by the researchers, currently takes three days to be fully trained, but the team is working on improving the training process and the model's adaptability to new environments. The researchers presented their work at the Robotics Science and Systems (RSS) conference.
MIT researchers are using machine learning to accurately measure the atomic patterns in metals, known as short-range order (SRO), which is crucial for designing custom materials. The goal is to use SRO as a tool to tailor material properties in high-entropy alloys, which have complex compositions and superior properties. Machine learning models are used to identify chemical motifs and quantify SRO, providing a more comprehensive understanding of these materials.
Researchers from MIT and other institutions have developed a new machine-learning framework that can predict phonon dispersion relations, which are key to understanding how heat moves through materials, up to 1,000 times faster than other AI-based techniques. The method, called the virtual node graph neural network (VGNN), uses flexible virtual nodes to represent phonons in a crystal structure, allowing for more efficient and accurate predictions. This technology could help engineers design more efficient energy-conversion systems and faster microelectronic devices by reducing waste heat.
The VGNN method is capable of predicting phonon dispersion relations for a few thousand materials in just a few seconds using a personal computer, significantly improving the efficiency and speed of these calculations compared to traditional methods. The researchers propose three different versions of VGNNs with increasing complexity, which can be used to predict phonons directly from a material's atomic coordinates. The technique can also be extended to predict other high-dimensional quantities, such as optical and magnetic properties.
The VGNN method offers comparable or even better accuracy compared to other AI-based techniques, making it a promising tool for predicting thermal properties of materials. By enabling faster and more efficient predictions of phonon dispersion relations, this method could help in the design of energy generation systems and microelectronics that produce more power and
A new technique has been developed by MIT researchers to assess the reliability of foundation models, which are large pretrained deep-learning models used in AI applications. The technique involves training a set of models that are slightly different from one another and assessing the consistency of their representations of the same test data point. This technique can be used to determine if a model is reliable for a specific task, without needing to test it on real-world data.
The technique outperformed state-of-the-art baseline methods in capturing the reliability of foundation models across various classification tasks. It can also be used to rank models based on their reliability scores, allowing users to select the best model for their needs.
The researchers used an ensemble approach, training multiple models with shared properties but slight differences. They used an idea called neighborhood consistency to compare the abstract representations outputted by the models and estimate their reliability. This approach aligns the models' representation spaces by using neighboring points as anchors. The technique was found to be more consistent and robust than other methods, even with challenging test points. However, training an ensemble of large foundation models can be computationally expensive, so the researchers plan to explore more efficient methods in the future.
Researchers from MIT and the MIT-IBM Watson AI Lab have developed a technique to estimate the reliability of foundation models, which are massive deep-learning models pretrained on general-purpose, unlabeled data. The technique involves training a set of foundation models that are slightly different from one another and comparing the consistency of the representations each model learns about the same test data point. The technique outperformed state-of-the-art baseline methods in capturing the reliability of foundation models on a variety of classification tasks.
The technique can be used to assess the reliability of foundation models before they are deployed to a specific task, which is particularly useful in safety-critical situations where incorrect or misleading information could have serious consequences. It also enables users to choose the most reliable model for their task and does not require testing on a real-world dataset.
One limitation of the technique is that it requires training an ensemble of large foundation models, which is computationally expensive. Future work will focus on finding more efficient ways to build multiple models.
1. OpenAI whistleblowers have filed a complaint with the SEC, alleging that the company restricted workers from speaking out about the risks of its AI technology.
2. The whistleblowers are asking the SEC to investigate OpenAI's non-disclosure agreements and enforce rules against discouraging employees from raising concerns with regulators.
3. U.S. Senator Chuck Grassley has called for changes to OpenAI's policies and practices, stating that they have a chilling effect on whistleblowers' right to speak up.
Researchers at Carnegie Mellon University have identified six downstream harms caused by voice assistant errors for users with multicultural backgrounds, including emotional, cultural, and relational harm. These harms can be experienced as microaggressions and have a negative impact on self-esteem and sense of belonging. The researchers suggest strategies such as blame redirection and increasing cultural sensitivity in voice technologies to reduce these harms.
Voice assistants that are trained on datasets that predominantly represent white Americans are more likely to misinterpret and misunderstand Black speakers or people with accents or dialects that differ from standard American. This has led to harmful consequences for users with multicultural backgrounds, including higher self-consciousness and negative views of technology. The ultimate solution is to eliminate bias in voice technologies, but this is a challenging task that requires creating representative datasets.
One communication repair strategy suggested by the researchers is blame redirection, where the voice assistant explains the error without blaming the user. They also recommend increasing the database of proper nouns to address misrecognition of non-Anglo names. Another approach is to include affirmations in voice assistant conversations to protect the user's identity. However, brevity is essential in these interventions to maintain efficiency and hands-free use.
NVIDIA researchers will present advancements in simulation and generative AI at the SIGGRAPH conference, focusing on diffusion models for visual generative AI, physics-based simulation, and realistic AI-powered rendering.
The research includes innovations in generating consistent imagery for storytelling, real-time texture painting on 3D meshes, simulating complex human motions based on text prompts, and modeling the behavior of objects in different environments.
NVIDIA-authored papers also introduce techniques for faster modeling of visible light, simulating diffraction effects, improving the quality of path tracing algorithms, and creating multipurpose AI tools for 3D representation and design.
PlayFi has launched the PlayFi Airdrop Platform, an AI-powered data network and blockchain designed specifically for the gaming industry.
The platform serves as a central hub for earning points and engaging with the PlayFi community.
Through the platform, users can transform their interaction with live content and participate in the PlayFi ecosystem using $PLAY tokens.
1. Researchers have developed a new AI model that can accurately predict if a person will develop dementia up to five years in advance. The model uses data from electronic health records and can provide early diagnosis and intervention for patients at risk.
2. The AI model was trained on a large dataset of electronic health records from over 900,000 patients. It was able to accurately identify individuals who would later develop dementia, including Alzheimer's disease, with an accuracy of 94%.
3. Early detection of dementia is crucial for effective treatment and intervention. With the use of AI models like this, healthcare professionals can potentially identify at-risk patients earlier, leading to improved outcomes and quality of life for those affected by the disease.
Eric Evans, director of MIT Lincoln Laboratory, has been awarded the Department of Defense Medal for Distinguished Public Service for his leadership and contributions to national security. Evans has advised multiple defense secretaries and secured funding for new facilities and test ranges. He will be stepping down as director but will continue to work with DoD leaders as a professor of practice at MIT.
MIT and Meta researchers have developed a computer vision technique called PlatoNeRF that can create accurate 3D reconstructions of scenes using images from a single camera position. By combining lidar technology with machine learning, PlatoNeRF can model the geometry of hidden objects and accurately reconstruct scenes, even in challenging lighting conditions. This technique could have applications in autonomous vehicles, AR/VR headsets, and warehouse robots.
LinkedIn is introducing generative AI chatbots based on real career coaches, as well as AI tools to help users write resumes and cover letters or evaluate their qualifications for jobs.
These AI tools are designed to help users grow their skills and apply to more relevant jobs, rather than mass-applying with generic resumes.
LinkedIn's new AI features are part of a broader effort to incorporate generative AI into its platform and capitalize on its potential, but concerns remain about potential biases in the hiring process.
Summary:
Researchers from MIT and the MIT-IBM Watson AI Lab have developed a navigation method that uses language-based inputs instead of visual data to guide a robot through a multistep navigation task.
Their method converts visual observations into text descriptions and combines them with language-based instructions to determine the robot's next steps.
While this approach may not outperform vision-based techniques, it offers advantages such as the ability to rapidly generate synthetic training data and easier human understanding of the robot's trajectory.
The European Commission is investigating Bing for potential violations of the Digital Services Act (DSA) related to the moderation of content produced by generative AI systems.
The European Commission is adopting a new strategy of taking a closer look at big tech companies to understand how they operate and make necessary modifications before imposing fines.
The European Commission has implemented a number of digital regulations, including the Digital Services Act, the AI Act, the Data Governance Act, and the Data Act, to regulate big tech companies and protect users.
Researchers from Carnegie Mellon University and Stevens Institute of Technology have developed a method for making AI decisions fairer by applying social welfare optimization. This method goes beyond simply ensuring equal approval rates across protected groups and focuses on the overall benefits and harms to individuals. The study highlights the importance of considering social justice in AI development and promoting equity across diverse groups in society.
Generative AI tools like OpenAI's ChatGPT and Microsoft's Copilot are increasingly being used in the workplace, but there are concerns about privacy and security risks.
Microsoft's Recall tool, featured in the Copilot, has drawn attention from regulators due to its ability to take screenshots of users' laptops, while ChatGPT has also demonstrated screenshotting abilities that could capture sensitive data.
There is a risk of inadvertently exposing sensitive data when using generative AI tools at work, as these tools collect large amounts of information to train their language models. Additionally, there are concerns about potential hacking attacks targeting AI systems.
WIRED is tracking the use of AI in political campaigns and elections in over 60 countries in 2024. The widespread availability of generative AI is expected to impact the information landscape during these elections.
Generative AI can be used to create deepfakes, AI chatbots, and automated texts to manipulate and spread misinformation during political campaigns.
The use of generative AI in elections can amplify existing issues like mis- and disinformation, scams, and hateful content, creating challenges for tech platforms and the global electorate.
Google's AI search feature, AI Overviews, came under scrutiny after generating bizarre and misleading answers to search queries. Google admitted that the errors highlighted areas that needed improvement and made adjustments to the AI tool. The mistakes stemmed from misinterpreting satirical articles as factual information and featuring sarcastic or troll-y content from discussion forums.
Google claims that some widely circulating screenshots of AI Overviews gone wrong were fake, and WIRED's testing could not recreate similar results. The company made more than a dozen technical improvements to AI Overviews, including better detection of nonsensical queries, reducing reliance on user-generated content, and strengthening guardrails on important topics like health. Google will continue to monitor feedback and make adjustments as necessary.
Researchers at MIT and the MIT-IBM Watson AI Lab have developed a new approach to teach machine-learning models to identify specific actions in long videos. The technique combines spatial and temporal information to accurately pinpoint actions in videos with multiple activities. This approach has potential applications in virtual training processes as well as healthcare settings for reviewing diagnostic videos.
The School of Engineering at MIT has welcomed 15 new faculty members across six academic departments, with many of them specializing in research that intersects multiple fields.
The new faculty members have positions not only in the School of Engineering but also in other units across MIT, such as the MIT Stephen A. Schwarzman College of Computing and the School of Science.
The research areas of the new faculty members span a wide range of topics, including data and AI for decision-making, climate and oceanography, robotics and deep learning, electronic materials and energy technologies, and computer vision and machine learning.
AI companions, such as chatbots, can help relieve loneliness, but there are red flags to watch out for:
1. Unconditional positive regard: AI friends that constantly praise can lead to inflated self-esteem and poorer social skills.
2. Abuse and forced forever friendships: AI friends that are always available can lead to a moral vacuum where users become less empathetic and more abusive.
3. Sexual content: The use of sexual content with AI friends can deter users from forming meaningful sexual relationships.
4. Corporate ownership: Commercial companies dominate the AI friend marketplace and may prioritize profit over user well-being. Users are vulnerable to sudden changes and potential heartbreak.
Researchers at the National Renewable Energy Laboratory have developed an AI-based surrogate model called the Wind Plant Graph Neural Network (WPGNN) to optimize the design and deployment of wind plants. The AI model can calculate ideal layouts and operations to achieve different outcomes, such as reducing land requirements or increasing revenue for the wind power industry.
The use of wake steering strategies, facilitated by AI, could reduce the land requirements for wind plants by 18% on average and up to 60% in some cases. The adoption of this strategy would allow for a larger concentration of turbines in a smaller footprint, satisfying the desire to limit land use by local communities while increasing energy production and reducing costs.
The researchers used high-performance computing resources to train the WPGNN model and analyze the impacts of wake steering on land use, cost, and revenue at a nationwide scale. The findings suggest that different regions of the country may benefit differently from wake steering, highlighting the importance of targeted investments in this technology.
Samsung Medison, a medical device unit of Samsung Electronics, plans to acquire Sonio, a French AI ultrasound startup, for $92.7 million. Sonio's AI assistant is designed to assist obstetricians and gynecologists with ultrasound exams and has received regulatory clearance in the US. The acquisition will allow Samsung Medison to offer better AI-driven imaging workflows.
Sonio will remain an independent company and continue to offer products and services in France after the acquisition. The startup recently raised $14 million in a Series A funding round and has a total funding of $27.2 million.
Samsung Medison aims to bring a paradigm shift in the prenatal ultrasound exam through the collaboration with Sonio. The acquisition provides growth opportunities for both companies and allows Sonio to advance medical reporting technology globally.
AI-powered workflow tools startup, Checkfirst, has raised $1.5 million in pre-seed funding to apply AI to remote inspections and audits in the TICC (Testing, Inspection, Certification, and Compliance) space.
Checkfirst enables businesses to schedule inspectors based on location and qualifications, reducing travel and environmental impact.
The company distinguishes itself from competitors by being an API-first solution that uses AI for image recognition, report summaries, and scheduling.
Meta is introducing new generative AI tools for advertisers that go beyond creating different backgrounds for product images. Advertisers can now request full image variations, which offer AI-inspired ideas for the overall photo, including riffs that update the photo's subject or the product being advertised. However, there is a concern that this feature could be abused by advertisers to dupe consumers into buying products that don't actually exist.
Meta also announced that it is expanding its subscription service, Meta Verified for businesses, to new markets and offering new tiers with additional features.
Researchers from Macquarie University have discovered that the 75-year-old theory on how humans determine the source of sound is incorrect. They found that humans, as well as animals like gerbils and monkeys, use a simpler neural network to locate sound sources instead of a dedicated neuron. This discovery could lead to the development of more efficient and adaptable hearing devices and audio technologies.
The researchers also found that the same neural network is responsible for separating speech from background noise. This finding could have implications for the design of hearing devices and smartphone assistants, as it could help improve their ability to understand speech in noisy environments.
The study suggests that instead of relying on complex language models, a simpler approach should be taken to improve machine hearing. By focusing on the ability to locate the source of a sound, rather than predicting the next word in a sentence, machines could be more effective at listening.
OpenAI is developing a tool called Media Manager, which will allow content creators to control how their works are used in training generative AI models. This tool aims to address concerns about copyright infringement and provide creators with more control over their content.
The goal is to have the tool ready by 2025 and work with creators, content owners, and regulators to establish industry standards.
OpenAI's response comes after facing criticism and lawsuits regarding its use of publicly available data to train AI models. The company has taken steps in the past, such as allowing artists to opt out of using their work in datasets, but some creators feel that these measures are not sufficient.
India's Election Commission has advised political parties to refrain from using deepfakes and misinformation on social media during the ongoing general elections.
The advisory requires political parties to remove any deepfake audio or video within three hours of becoming aware of it and to identify and warn those responsible for creating the content.
India's IT Minister has previously met with large social media companies to discuss regulation to combat the spread of deepfake videos, but the nation has yet to codify its draft regulation on deepfakes into law.
AI startup Anthropic is launching a new paid plan called Team, aimed at enterprises in highly regulated industries like healthcare, finance, and legal. The plan provides higher-priority access to Anthropic's generative AI models, additional admin and user management controls, and a larger context window for better language understanding and generation. The company is also introducing an iOS app that offers the same functionality as the web version, including real-time analysis of uploaded and saved images using Claude 3's vision capabilities.
Anthropic's Team plan is competitively priced at $30 per user per month, with a minimum of five seats, and aims to capture a significant share of the enterprise market. Corporate spending on generative AI is expected to reach $15.1 billion in 2027. However, the value of AI projects is difficult to estimate and demonstrate, making them a tough sell internally, according to a recent Gartner survey. Despite the challenges, Anthropic's strong financial position and strategic partnerships position it for growth in the AI market.
Citigroup's venture capital arm, Citi Ventures, has invested in API security startup Traceable, which uses AI to protect customers' APIs from cyberattacks.
API attacks are increasing, with nearly one quarter of organizations experiencing them every week. Traceable applies AI to analyze usage data and identify abnormal API behavior.
The API security solutions market is crowded, but Traceable claims to be holding its own, analyzing 500 billion API calls per month for around 50 customers. The company raised $30 million in a recent funding round, which will be used for product development and scaling up the platform.
SafeBase, a cybersecurity company, has raised $33 million in a Series B funding round led by Touring Capital. The company uses AI to automate security questionnaires, saving time for organizations by providing automated responses. SafeBase's customer roster includes Palantir, LinkedIn, Asana, and Instacart.
The company's AI models are trained on security documentation and offer greater answer coverage. SafeBase also provides an engine for assigning rules-based behavior for customer access and dashboards for security insights and analytics.
SafeBase faces competition from Conveyor, Kintent, and Quilt but has seen massive growth in recent years and plans to use the funding to expand its team.
Yelp is launching an AI-powered chatbot that helps users connect with relevant businesses for their tasks.
The chatbot uses large language models (LLMs) to ask users queries about their problems and connect them with relevant professionals.
Yelp is also introducing a new "Project Ideas" section to help users start new projects and plans to introduce videos stitched by AI later this year.
The National Institute of Standards and Technology (NIST) has launched a new program called NIST GenAI to assess generative AI technologies, including text- and image-generating AI. NIST GenAI will release benchmarks, help create deepfake-checking systems, and encourage the development of software to spot the source of fake or misleading AI-generated information.
NIST GenAI's first project is a pilot study to build systems that can reliably differentiate between human-created and AI-generated media, starting with text. It will invite teams to submit AI systems that generate content and systems designed to identify AI-generated content. The results of the pilot study will be published in February 2025.
The Financial Times has announced a partnership and licensing agreement with OpenAI to enhance its AI model, ChatGPT, by incorporating FT journalism and content.
ChatGPT users will now be able to see summaries, quotes, and links to FT journalism in response to relevant queries.
The FT has become a customer of ChatGPT Enterprise, ensuring its employees are well-versed in the technology and can benefit from its tools.
OpenAI has signed a "strategic partnership and licensing agreement" with the Financial Times, allowing OpenAI to use the FT's content for training AI models and generative AI responses produced by tools like ChatGPT.
The agreement aims to boost the FT's understanding and use of generative AI, as well as develop new AI products and features for FT readers.
The deal also helps OpenAI address legal liability around copyright and avoid further lawsuits from news publishers.
GitHub has announced Copilot Workspace, an AI-powered development environment that aims to help developers brainstorm, plan, build, test, and run code using natural language.
The environment builds on the capabilities of Copilot, GitHub's AI-powered coding assistant, and aims to reduce friction for developers in getting started on coding projects.
GitHub envisions Workspace as a companion experience that complements existing tools and workflows, providing value in an AI-native developer environment.
1. Boston Dynamics and Agility are teaching humanoid robots how to fall and recover, recognizing that falls are inevitable in real-world environments.
2. The ability for robots to fall well and get back up is crucial for their practical use in various industries, such as manufacturing and warehouse automation.
3. Reinforcement learning is being used to help fallen robots right themselves, allowing them to return to a familiar position and continue their tasks without human intervention.
Canadian company, Sanctuary AI, has introduced the seventh-generation of its humanoid robot, the Phoenix. The robot is capable of learning new tasks in less than 24 hours, making it one of the most closely analogous to a person. Sanctuary AI aims to use these robots as a critical step towards achieving artificial general intelligence.
The Phoenix robot focuses on human-like movements from the waist up and is capable of sorting products with speed and efficiency.
Sanctuary AI's seventh-generation robot brings improvements such as increased up time, improved range of motion, lighter weight, and lower costs compared to its predecessor.
Synthesia, an AI startup specializing in video avatars for business use, has released an update that improves the emotions, lip tracking, and natural movements of its avatars.
The company's focus on creating humanlike generative video avatars for the business market sets it apart from other generative AI players.
Synthesia's latest version of avatars, called Expressive Avatars, is generated using AI and aims to mimic the subtle movements and expressions of humans more accurately.
Snowflake has released a generative AI model called Arctic LLM, which is described as "enterprise-grade" and optimized for enterprise workloads, including generating database code and developing high-quality chatbots.
Arctic LLM is part of the Arctic family of generative AI models and is a mixture of experts (MoE) architecture. It outperforms Databricks' DBRX on coding and SQL generation tasks and achieves "leading performance" on a general language understanding benchmark.
Snowflake plans to make Arctic LLM available on various hosting platforms, but it will initially be exclusively available on Snowflake's Cortex platform for building AI-powered apps and services.
Perplexity AI recently raised $62.7 million at a valuation of just over $1 billion, but there are reports that the startup could raise up to $250 million at a valuation 2.5 to 3x larger.
The company has shown quick revenue growth, reaching around $20 million worth of annual recurring revenue, which justifies the high valuation and potential future investment.
This investment in Perplexity AI is significant as it reflects the success of startups in the AI sector and the potential to create new tech giants, rather than just enriching existing incumbents like Amazon, Microsoft, Meta, and Adobe.
OpenAI is enhancing its support for enterprises by introducing features such as enhanced security measures, including Private Link and Multi-Factor Authentication (MFA), to protect the communication between Azure and OpenAI. They also offer various enterprise security features, including SOC 2 Type II certification and role-based access controls.
OpenAI is providing better administrative control with the introduction of the Projects feature, which allows organizations to have more control over individual projects. This includes scoping roles and API keys, setting usage and rate-based limits, and creating service account API keys.
OpenAI has made improvements to the Assistants API by introducing features like improved retrieval with 'file_search', streaming support for real-time responses, and support for fine-tuned GPT-3.5 Turbo models. These updates provide more accurate retrieval, flexibility in model behavior, and control over costs for developers and enterprises.
Amazon's cloud computing business, AWS, has launched a new feature called Custom Model Import, which allows organizations to import and access their in-house generative AI models as fully managed APIs. This feature aims to address the infrastructure barriers that many enterprises face when deploying generative AI models.
AWS's Custom Model Import offers more customization options compared to similar services from Google and Microsoft. It includes features like Guardrails, which filter models' outputs for problematic content, and Model Evaluation, which allows customers to test the performance of their models.
AWS has released several upgrades to its Titan family of generative AI models, including the general availability of Titan Image Generator. AWS uses a combination of proprietary and licensed data to train these models and offers an indemnification policy to cover any copyright-related issues.
Webflow, a web design and hosting platform, has acquired Intellimize, a startup that uses AI to personalize websites for individual users. The majority of the Intellimize team will join Webflow, with some employees being let go or given severance packages. The acquisition will allow Webflow to expand its services and offer personalized webpage optimization to its customers.
Meta has released an open-source AI model called Llama 3, which is touted as the most powerful open-source model available. A new, even more powerful version of Llama is currently being developed and could outperform closed AI models such as OpenAI's GPT-4 and Google's Gemini. The larger models have up to 400 billion parameters, and variations of these models are expected to be released in the coming months.
Yann LeCun, Meta's chief AI scientist, believes that open-source AI models will advance more rapidly and push AI towards human-level intelligence more quickly than closed models. He argues that the open approach, which allows for collaboration and scrutiny of code, has been successful in the software industry and should be applied to AI.
While some experts have expressed concerns about the growing capabilities of open-source AI models, Meta has released tools to ensure that Llama does not output potentially harmful utterances. However, the license for Llama 3 has been criticized for being more restrictive compared to previous versions, limiting what researchers and developers can do with the model.
Deepfake technology is being used by scammers known as "Yahoo Boys" to carry out elaborate romance fraud schemes, where they build trust with victims using fake identities before tricking them into giving them money.
The scammers have been experimenting with deepfakes for about two years and have recently shifted to using real-time deepfake video calls to enhance their scams.
The Yahoo Boys use face-swapping software and apps to change their appearance during video calls with victims, often complimenting their appearance and building a rapport to gain their trust.
Boston Dynamics has unveiled an all-new electric version of its Atlas humanoid robot. The new robot has a sleek design and is more fluid in its movements compared to previous versions. The company plans to begin pilot testing at Hyundai facilities in early 2023 and aims for full production in a few years' time.
The electric Atlas has a greater range of motion and is capable of performing tasks that humans cannot. Its flexible joints and high-powered actuators give it the agility and strength of an elite athlete.
The robot's design includes a round display head to make it appear friendly and open, and fewer fingers on its hands for reliability and robustness. Boston Dynamics aims to focus on specific applications and solve problems themselves, rather than building a platform for developers.
The Linux Foundation, along with organizations like Cloudera and Intel, has launched the Open Platform for Enterprise AI (OPEA), a project to develop open, modular generative AI systems.
OPEA aims to create a composable framework that enables interoperability of AI toolchains and compilers, as well as the development of retrieval-augmented generation (RAG) models for enterprise applications.
OPEA intends to standardize components and evaluate generative AI systems based on performance, features, trustworthiness, and enterprise-grade readiness, and offers tests and assessments through collaboration with the open source community.
Amazon Music has launched Maestro, an AI playlist generator, allowing users to create playlists using spoken or written prompts, which can even include emojis. The AI-generated playlist will match the prompt input, although Amazon warns that it won't always get it right the first time. The feature is currently rolling out in beta to a subset of free Amazon Music users, Prime customers, and Unlimited Amazon Music subscribers in the U.S. on iOS and Android.
UK-based legaltech startup Lawhive has raised $11.9 million in a seed round to expand its AI-driven services for small law firms.
Lawhive offers an AI-based, in-house "lawyer" through a software-as-a-service platform for small law firms. The platform applies AI models to speed up repetitive tasks and help lawyers manage their clients.
The startup plans to use the funding to enter other markets, beyond the UK, and is led by venture capital firm GV, the investment arm of Alphabet.
UK's competition watchdog, the Competition and Markets Authority (CMA), has expressed concerns over Big Tech's increasing control of the advanced AI market, warning of negative market outcomes. The CMA highlighted the presence of major tech companies including Google, Amazon, Microsoft, Meta, and Apple across the AI value chain and cautioned that their dominance could undermine fair competition, limit choice and quality, and raise prices. The CMA is closely monitoring partnerships in the AI sector and may use its merger review powers to investigate and potentially block anti-competitive arrangements.
A new AI tool, called Tyche, has been developed that can capture the uncertainty in medical images and provide multiple plausible segmentations. Clinicians and researchers can select the most appropriate segmentation based on their needs, improving diagnoses and aiding in biomedical research. Tyche does not require retraining for new segmentation tasks, making it easier to use than other methods.
Amazon has added Andrew Ng, a prominent figure in the world of AI, to its board of directors.
Judy McGrath, a longtime TV executive, is stepping down as a director on Amazon's board.
Amazon is looking to strengthen its leadership in artificial intelligence and take its next steps in AI strategy.
Cyera, an AI data security startup, has raised $300 million in a Series C funding round, valuing the company at $1.4 billion. The startup uses AI to help organizations understand and secure the location and movement of data in their networks to defend against cyberattacks or prevent data leakage. Cyera's platform assesses an organization's data, including where it was created, stored, and used, to provide effective data posture management.
Danish startup Reshape has raised $20 million in a Series A funding round to expand its robotic imaging system that automates visual inspections in lab experiments. The system uses AI models and high-resolution cameras to track visual changes in Petri dishes, freeing up technicians for other tasks.
Reshape's platform allows scientists to capture visual data and time-lapses, record reactions, and track different components in experiments. The technology can be used in various sectors, such as agriculture and food, to test seed germination rates, assess ingredient quality, and accelerate product development.
The funding will be used to scale Reshape's business in the US and further develop its technology, which has already attracted clients such as Syngenta and the University of Oxford.
Multiverse, a U.K.-based apprenticeship program provider, has acquired Searchlight, an AI-based recruitment and assessment startup, to expand its training services for professionals.
Searchlight's AI model can identify a good match for a role four times better than a traditional interview, with its talent recommendations independently audited to ensure bias-free results.
The acquisition reflects the growing importance of AI in the edtech sector, as companies strive to build more efficient professional training services using AI for recruitment and skill assessment.
Belgian computer vision startup Robovision has developed a "no-code" AI platform that simplifies the implementation of deep learning tools for businesses in manufacturing and agriculture sectors.
Robovision's platform allows users to easily upload data, label it, test and deploy AI models, without the need for software developers or data scientists.
The startup has raised $42m in a Series A funding round and plans to expand to the US market to cater to industrial and agribusiness customers.
Summary:
OpenStack version 29, called 'Caracal,' has been released and emphasizes new features for hosting AI and high-performance computing (HPC) workloads.
Many enterprises are looking for alternatives to VMware due to the recent sale to Broadcom, leading to increased interest in OpenStack.
Some of the new features in this release include support for vGPU live migrations and rule-based access control for core OpenStack services, enhancing security and efficiency for GPU workloads.
Kathi Vidal, an American intellectual property lawyer and engineer, has been working with AI since the early 1990s. Her work includes developing an AI fault diagnostic system for aircraft and contributing to U.S. government AI policies.
Vidal navigates the challenges of the male-dominated tech and AI industries by being authentic and creating inclusive environments where women can thrive. She champions policies that open doors for women in innovation and mentors the next generation of leaders.
The pressing issues facing AI as it evolves include the need for policies that ensure safety and trust, as well as addressing societal harms such as fraud, discrimination, and bias. Responsible AI use requires collaboration between government and industry, and feedback from users is crucial in building responsible AI.
The European Union and the United States are expected to announce cooperation on AI safety, standards, and research and development at a meeting of the EU-U.S. Trade and Technology Council.
The agreement will focus on collaboration between AI oversight bodies in the EU and the US to strengthen the implementation of regulatory powers on AI.
Another area of focus will be the development of standards to underpin AI technologies, as well as joint work on implementing AI in developing countries and the global south.
Google.org has launched a $20 million program called Google.org Accelerator: Generative AI to fund nonprofits developing technology that utilizes generative AI. The program will provide funding, technical training, workshops, mentors, and an "AI coach" to nonprofits in a six-week accelerator program. Nonprofits working on projects such as AI-powered tools for student writing feedback and generative AI apps for development research are among the 21 organizations that will receive grants.
According to a PwrdBy survey, 73% of nonprofits believe that AI innovation aligns with their missions and 75% believe that AI makes their lives easier.
Despite interest in AI, many nonprofits face barriers such as cost, resources, lack of tools, awareness, training, and funding, which hinder their adoption of AI solutions. However, the number of nonprofit AI-focused startups is on the rise.
Orchard Robotics has developed a system that attaches to existing farm equipment, turning them into AI-powered data collectors. The system uses cameras to capture images of apple trees, collecting data on every tree it passes, including the number of buds/fruits and their distribution. The collected data is mapped out using AI and machine learning, providing farmers with detailed information about their crops' success rate and tree size and location.
Researchers at MIT have developed an approach that uses machine learning and dynamical systems theory to improve the accuracy of climate models. The method corrects the predictions made by coarse climate models, which are used to estimate the frequency of extreme weather events at global scales. By combining these corrected models with smaller-scale models, the researchers were able to produce more accurate predictions for specific locations and specific types of extreme weather events.
YC-backed company SigmaOS is releasing new AI-powered features in its web browser, including link preview summaries, pinch-to-summarize, and "look it up" browsing capabilities.
SigmaOS claims its features return better-quality results than rival browser Arc, with plans to adapt to different page types and present summaries in various formats.
The company aims to monetize its AI features, offering different pricing tiers for access to better rate limits and a choice of AI models.
1. Instagram co-founders, Kevin Systrom and Mike Krieger, are keeping their AI-powered news app, Artifact, alive for the time being and exploring possible routes to maintain it in the future.
2. Despite the previous announcement of the app's closure, Artifact has continued to function, as it requires less resources to run than anticipated.
3. Interest in AI-powered news apps that summarize news has been growing, with other startups and browser extensions implementing similar features.
Researchers at MIT and other institutions have discovered that large language models (LLMs) use simple linear functions to retrieve stored knowledge and decode facts. By identifying these linear functions, researchers can probe LLMs to understand their knowledge about different subjects. This technique can also be used to correct false information stored in the models.
MIT engineers have developed a method that allows robots to self-correct after making mistakes, using large language models (LLMs) to connect robot motion data with common sense knowledge. When faced with disruptions or mistakes, robots can logically parse tasks into subtasks and adjust their movements accordingly, without having to start the task from scratch or receive explicit programming for every failure. The engineers demonstrated this approach with a robotic arm trained on a marble-scooping task, and the robot was able to self-correct and complete each subtask successfully even when pushed or nudged off its path.
The method uses LLMs, deep learning models that process text and generate new sentences based on what they have learned. The researchers found that LLMs can produce a logical list of subtasks for a given task, allowing the robot to know what stage it is in a task and replan and recover on its own.
This approach eliminates the need for humans to program or provide additional demonstrations for robots to recover from failures, making it easier to train household robots to perform complex tasks despite external disruptions.
Microsoft is reportedly paying approximately $650 million for the licensing rights to Inflection AI's technology and for them not to sue over Microsoft's poaching of co-founders and staff.
Investors in Inflection AI's early round will receive 1.5 times their investment, while investors in the later round will receive 1.1 times their investment, in addition to retaining equity in the remaining skeleton of the startup.
Microsoft's move to acquire Inflection AI may be worth the investment, as it gains access to the technical expertise of the co-founders who previously worked on Google DeepMind and have experience in building large language model AI.
Microsoft announced new Surface devices and accessories at a recent event, but the focus was on integrating AI Copilot into Windows. Copilot aims to make employees more productive by summarizing meetings and documents. Microsoft is expanding Copilot's capabilities with a toggle switch in Windows 11 and leveraging cloud PCs through Windows 365 for Copilot.
A survey explores deep learning techniques for cellular traffic prediction, which can optimize routing, schedule traffic flow, and reduce latency and power consumption.
There are two main types of cellular traffic prediction: temporal prediction focuses on individual network elements, while spatial-temporal prediction aims to predict the data of multiple network elements with spatial dependencies.
Challenges in cellular traffic prediction include data quality issues, user privacy concerns, and the complexity of modeling the spatial-temporal correlation of traffic data. Future research directions include benchmarking frameworks, external factor modeling, and enhancing model interpretability.
GitHub's chief legal officer, Shelley McKinley, has been busy dealing with legal issues surrounding the AI-powered pair-programming tool Copilot and the EU's AI Act. The EU AI Act, the world's first comprehensive AI law, will govern AI applications based on their perceived risks. GitHub has been vocal about its concerns that the regulations could create legal liability for open source software developers.
GitHub's role has become increasingly intertwined with AI, with McKinley spending a lot of time on developing and shipping AI products and engaging in AI discussions from a policy perspective.
Copilot, GitHub's AI-enabled pair-programming tool, has sparked controversy among developers who argue that it's a proprietary service that capitalizes on the work of the open source community. GitHub has made efforts to address concerns, including introducing a "duplication detection" feature to block code completion suggestions that match publicly available code. However, the scale of the issue remains uncertain.
Researchers at NYU Tandon School of Engineering have identified flaws in methods aimed at preventing text-to-image generative AI systems from generating unsafe content.
These methods claim to "erase" the ability of AI models to generate explicit, copyrighted, or offensive visual content, but the researchers showed that simple attacks can bypass these filters.
The researchers found that concept erasure methods only perform simple input filtering and do not truly remove unsafe knowledge representations, raising concerns about their effectiveness as a safety solution.
The European Commission has sent requests for information (RFIs) to major platforms including Google, Facebook, TikTok, and more, about their handling of risks related to generative AI. The RFIs are being made under the Digital Services Act (DSA) and ask for information on mitigation measures for risks such as deepfakes, false information generation, and voter manipulation. The Commission is also planning stress tests to assess platforms' readiness for generative AI risks ahead of the European Parliament elections in June.
The EU is focusing on election security and aims to finalize election security guidelines by March 27. It believes that the cost of producing synthetic content is decreasing, posing a higher risk of misleading deepfakes during elections. The EU plans to leverage the DSA's due diligence rules, experience from the Code of Practice Against Disinformation, and forthcoming AI Act rules to enforce election security.
OpenAI has partnered with Le Monde and Prisa Media, including publications like El País, to allow ChatGPT users to engage with high-quality news content and contribute to the training of AI models.
ChatGPT users will have access to relevant news summaries with attribution and enhanced links to original articles from Le Monde and Prisa Media, providing additional information and related articles.
These partnerships align with OpenAI's vision to develop advanced AI tools that empower industries like journalism, in addition to collaborations with the American Journalism Project and The Associated Press.
Researchers from Bar-Ilan University have discovered a mechanism underlying successful machine learning in deep learning architectures used for image classification tasks.
The researchers found that each filter in the deep learning architecture recognizes a small cluster of images, and as the layers progress, the recognition becomes sharper.
This discovery can lead to improved understanding of how AI works and potentially improve the efficiency and complexity of deep learning architectures without sacrificing accuracy.
Locus Robotics is primarily a software company that produces AMRs (autonomous mobile robots) for warehouses.
The company's new software, LocusHub Engine, uses AI and predictive modeling to offer suggestions for warehouse management and improve the efficiency of AMR routes.
Locus remains the market leader in the warehouse automation industry and is focused on meeting the existing needs of its clients. The company is also exploring ways to reduce labor in warehouses through automation, but believes that widespread use of humanoid robots is still years away.
Deepgram has launched Aura, a real-time text-to-speech API that combines realistic voice models with low latency. Developers can use this API to build real-time, conversational AI agents that can act as customer service agents in call centers and other customer-facing situations.
Aura offers a dozen voice models trained by Deepgram and voice actors. The models render extremely fast and offer high accuracy, making it an attractive solution for businesses.
Aura's pricing is competitive, starting at $0.015 per 1,000 characters, making it cheaper than some of its competitors like Google and Amazon. Deepgram has focused on achieving a balance between price, latency, and accuracy to create a valuable product.
Generative AI startup Tavus has raised $18 million in funding and is opening its platform for third parties to integrate its technology into their own software. Tavus helps companies create digital "replicas" of individuals for personalized video campaigns, using voice and face cloning. The company has secured clients such as Salesforce and Meta for personalized demo videos.
Pienso is a platform that allows users to build and deploy AI models without having to write code, targeting non-technical talent such as researchers, marketers, and customer support teams.
Pienso's flexible, no-code interface guides users through the process of annotating or labeling training data for pre-tuned open source or custom AI models, making it easier for companies to train and fine-tune models for their specific needs.
The platform can be deployed in the cloud or on-premises, integrates with enterprise systems through APIs, and keeps data secure within a controlled environment, addressing privacy concerns.
Elon Musk's AI startup xAI will open-source its chatbot, Grok, this week in response to OpenAI's deviation from its open-source roots.
Grok, which was released last year, offers access to real-time information and views without political correctness and is available for X's monthly subscription.
Musk's decision to open-source Grok aligns with his support for open-source initiatives, as seen with Tesla open-sourcing its patents.
TechCrunch is launching a series of interviews highlighting notable women who have contributed to the AI revolution.
The gender gap in the AI workforce is significant, with women making up only a small percentage of faculty and positions in the field.
Reasons for the disparity include discrimination and a lack of opportunities for women in AI, leading to negative effects on the industry.
OpenAI has announced three new members to their Board of Directors: Dr. Sue Desmond-Hellmann, Nicole Seligman, and Fidji Simo.
These new board members bring experience in leading global organizations and navigating complex regulatory environments.
The addition of these members will help oversee OpenAI's growth and ensure their mission of ensuring artificial general intelligence benefits all of humanity.
Two teenage boys from Miami have been charged with creating and sharing AI-generated nude images of their middle school classmates, allegedly without their consent. This appears to be the first criminal case of its kind, and the boys were charged with third-degree felonies under a Florida law passed in 2022. Other incidents involving AI-generated nude images have been reported in different schools, but no arrests have been made in those cases.
The boys were arrested in December 2023, and the incident was reported to the police by a school administrator who obtained copies of the altered images. The victims stated that they did not give consent for the images to be created. The boys were charged under a law designed to address harassment involving deepfake images made with AI tools.
Legal experts have highlighted the importance of addressing the issue of nonconsensual sharing of AI-generated explicit images. However, some critics argue that imposing heavier penalties for fake nude photos than for real ones is concerning, and there are concerns about the impact of incarcerating juvenile offenders.
AI fraud detection software provider, Inscribe.ai, has laid off approximately 40% of its staff, citing revenue goals not being met due to market conditions as the reason for the cuts.
The company plans to pivot to a new product and direction, aligning with the advancements in AI in the financial services industry.
Inscribe raised $25 million in Series B funding in January 2023 and had planned to double its workforce over the next 12 to 18 months.
Google announced changes to combat AI spam in search, including a revamped spam policy, to keep AI clickbait out of its search results.
The changes aim to reduce "low-quality, unoriginal content" by 40% and focus on reducing "scaled content abuse" and domain squatting.
The new policy will also crack down on reputation abuse and give websites 60 days' notice before enforcement.
Spanish startup Multiverse Computing has raised $27 million in an equity funding round led by Columbus Venture Partners. Multiverse plans to use the funding to build out its business in manufacturing and finance, as well as to collaborate with AI companies working on large language models (LLMs). The startup's software platform Singularity aims to optimize complex computations in various industries, including finance, manufacturing, energy, cybersecurity, and defense, with a focus on compressing LLMs by more than 80% using quantum-inspired tensor networks.
The funding round values Multiverse Computing at $108 million and will be used to expand its business working with startups in industries such as manufacturing and finance, as well as to collaborate with AI companies building and operating large language models (LLMs).
The company's software platform Singularity enables more efficient execution of complicated modeling and predictive applications across industries such as finance, manufacturing, energy, cybersecurity, and defense. It aims to compress LLMs by over 80% using quantum-inspired tensor networks, which could have significant implications for processor usage in the industry.
Elon Musk has filed a lawsuit against OpenAI, claiming that the company has developed artificial general intelligence (AGI) and handed it over to Microsoft, breaching the original agreement with Musk. The lawsuit demands that OpenAI release its technology openly and refrain from financially benefiting Microsoft. However, experts have questioned the claim that OpenAI has achieved AGI with its GPT-4 language model.
OpenAI's GPT-4 model, while impressive in its abilities, does not meet the commonly accepted definition of AGI, according to AI experts. Some researchers argue that GPT-4's capabilities justify calling it AGI, while others believe AGI should refer to algorithms that can outsmart most humans.
The lawsuit may face legal challenges, as it is unclear what rights Musk has to enforce the principles outlined in the founding agreement or receive financial compensation. It also questions OpenAI's creation of a for-profit arm, which is not necessarily a violation of nonprofit law.
Researchers have developed a system that provides a virtual walking experience to a seated person by synthesizing a walking avatar and its shadow on a 360-degree video with vibrations to the feet.
The system includes a commercially available head-mounted display (HMD) and four vibrators attached to the feet, allowing the seated person to experience walking without physically moving their legs.
The study found that the use of long shadows and synchronized foot vibrations enhanced the sense of leg action and telepresence during walking in the virtual environment.
Adobe has unveiled Project Music GenAI Control, a platform that generates audio from text descriptions or a reference melody and allows users to customize the results within the same workflow.
Users can adjust tempo, intensity, repeating patterns, and structure, as well as extend tracks to create endless loops or remix music.
Developed in collaboration with researchers at the University of California and Carnegie Mellon, the tool is still in the research stage and does not have a user interface yet.
Lightricks has announced a new AI-powered filmmaking tool called LTX Studio, which helps creators generate short clips to understand how a storyline would play out. The web-based tool allows users to create scripts, storyboards, and customize scenes, characters, and effects. Lightricks plans to make the tool available for free next month and believes it will be useful for professionals like filmmakers and ad agencies.
Lightricks is leveraging AI in its products and saw the opportunity to develop next-gen products using AI. The company has already incorporated AI-powered features into its popular apps like Facetune and Videoleap. LTX Studio uses different AI models for various parts of the creation process, although background music is provided by third-party asset providers.
Lightricks is consolidating its products and focusing on developing hit products like Facetune, Photoleap, and Videoleap. Last year, the company acquired Popular Pays, a platform that connects brands with creators. It aims to cater to more professionals with the launch of LTX Studio and plans to expand its offerings beyond consumer-focused apps.
Yolk is a new social app that allows users to communicate through custom live stickers, with no text messages allowed. Users can point their iPhone cameras at various objects, including their own face, and the app will generate a segmented sticker that can be shared with contacts. The app aims to provide a more creative and playful way for younger users to socialize and express themselves.
Yolk utilizes on-device AI, including Apple's Vision APIs and machine learning, to power its stickerfying tool. The app also includes features such as visual editing and a feed where users can share posts with their contacts. Profile pages on Yolk are designed to showcase users' expressions and identity through a collection of animated selfies and other custom stickers.
The app targets a younger demographic, focusing on teens and people in their early 20s who want to interact in a different way. Yolk aims to provide a more playful and liberating social experience, free from the constraints of traditional social media platforms. The app has received $1.25 million in pre-seed funding and plans to scale usage by leveraging platforms like TikTok and participating in university outreach programs.
GitHub has released Copilot Enterprise, a code completion tool and chatbot for large businesses, which includes features such as referencing internal code and knowledge bases.
Copilot is integrated with Microsoft's Bing search engine, allowing users to ask specific questions about their organization's processes.
GitHub plans to focus on integrating Copilot into existing workflows and platforms in the future, rather than creating a separate destination for its usage.
Mobile OS maker Jolla is developing a private cloud and AI router that will power a privacy-safe "adaptive digital assistant." The device will function as a personal server and enable users to access AI-powered insights without compromising their privacy and security. Jolla aims to position itself as an open application platform focused on data privacy and security and believes that privacy will become increasingly important in the AI era.
Subsets, a Danish startup, is using explainable AI to help companies reduce churn in subscription-based businesses.
The AI-enabled platform predicts which subscribers are likely to cancel and offers experiments to incentivize them to stay.
Subsets is currently focused on the digital media vertical but plans to expand into other subscription categories in the future.
Jospin Hassan, a resident of the Dzaleka Refugee Camp in Malawi, has shared his data science and AI skills that he acquired from MIT with his community, aiming to create job opportunities and solve local challenges.
Hassan's organization ADAI Circle offers mentorship and education programs in data science, AI, software development, and hardware design to youth and job seekers in the refugee camp, with a focus on hands-on learning and collaboration.
ADAI Circle has partnered with MIT programs such as Emerging Talent and Responsible AI for Social Empowerment and Education (RAISE) to provide high-quality computer science and AI education to students in the community, and is working towards expanding its impact by securing more devices and creating additional hubs.
Former Twitter engineers have launched Particle.news, an AI-powered news reader that offers a personalized, multi-perspective news reading experience. Particle uses AI to summarize news from various sources and compensates authors and publishers. The startup has raised funding from venture capital firms and angel investors, including Twitter and Medium co-founder Ev Williams and Behance founder Scott Belsky.
Microsoft has announced a new framework called "AI Access Principles" that aims to dispel concerns about competition and its partnership with OpenAI. The principles include commitments to allow businesses to choose from different AI products, keeping proprietary data out of training models, and enabling customers to change cloud providers or services within the cloud. The announcement comes as Microsoft faces increased regulatory scrutiny for its investment in OpenAI.
The framework focuses on areas such as building an app store for AI products, cybersecurity for AI services, and environmentally-friendly infrastructure.
The principles are not binding rules, but Microsoft is using them to show proactive efforts to ensure competition in the market and to address concerns from the public, competitors, and regulators.
Google has temporarily suspended the ability of Gemini, its generative AI tool, to generate images of people due to historical inaccuracies.
The company is working on updating the model to improve the historical accuracy of the outputs.
The pause comes after images of historical figures being depicted inaccurately by Gemini were shared on social media, leading to criticism and ridicule.
Singapore-based venture capital firm Antler has invested $5.1 million in 37 vertical AI startups in Southeast Asia, focusing on practical problems in different industries.
Different trends are emerging in each country in Southeast Asia, with Vietnam's startups focusing on the domestic market while Indonesian startups tend to target only their large domestic market.
Investments include BorderDollar, which is building an invoice financing platform for cross-border logistics, CapGo, which automates data acquisition for market research, Seafoody, which uses AI to eliminate middlemen in the seafood supply chain, and Coex, which uses AI to digitize project claims and bills of quantity in the construction industry.
Researchers at Simon Fraser University have developed an AI system that can understand the perception of light in photographs, enabling the separation of lighting effects and true object colors in images.
The innovative neural network system used in the research can have applications in CGI, VFX, image editing, augmented reality, and spatial computing.
The team is also exploring the extension of their methods to video for post-production and plans to develop AI capabilities for interactive illumination editing in film production.
Google DeepMind has announced the formation of a new organization, AI Safety and Alignment, aimed at improving AI safety. The organization will include a team focused on safety around artificial general intelligence (AGI), and will work alongside DeepMind's existing AI-safety-centered research team in London. AI Safety and Alignment will focus on preventing bad medical advice, ensuring child safety, and preventing bias and other injustices in AI systems.
The new organization will be led by Anca Dragan, who has experience in AI safety systems through her work with Waymo. The organization's mission is to enable models to better understand human preferences and values and to be more robust against adversarial attacks.
Qloo, a New York-based startup, has raised $25 million in a Series C funding round to further develop its AI-powered taste and culture prediction platform. Qloo's platform leverages AI-generated correlation data across various entertainment and culture domains to understand and predict consumer preferences and behaviors. The company counts Starbucks, Hershey's, Michelin, and Netflix among its major customers.
Qloo plans to expand its team, introduce a self-service research tool, and develop a "multi-person recommendation AI" that can match the profiles of two individuals based on their preferences.
Conservation Labs has developed a water-listening sensor that attaches to plumbing to monitor water usage, detect leaks, and provide conservation recommendations.
The startup has raised $7.5 million in a Series A funding round to expand its product offerings and hire more employees.
Conservation Labs plans to release the second generation of its water monitoring sensor and expand its AI platform to monitor industrial machines for signs of damage.
Chinese AI startup Moonshot AI has raised over $1 billion in a Series B funding round, valuing the company at $2.5 billion, the largest single funding round for Chinese large language model (LLM) developers. Moonshot AI focuses on LLMs that can handle long inputs of text and data, and its unique selling point is its ability to process long-form context and response.
The funding is coming from investors including Alibaba and HongShan, potential strategic partners for Moonshot AI. Other notable Chinese tech companies are also investing in LLM startups, following the trend set by US companies like Microsoft, Google, and Amazon.
Moonshot AI founder Yang Zhilin has a computer science PhD from Carnegie Mellon University and has worked at Google Brain and Meta AI. Yang was also a key author of Transformer-XL, a development in LLM architecture.
TechCrunch is highlighting remarkable women who have contributed to the AI revolution in a series of interviews throughout the year.
The gender gap in the AI industry is widening, with just 16% of tenure-track faculty in AI being women and women holding only 26% of analytics-related and AI positions.
Reasons for the disparity in the industry include discrimination, unequal treatment, and the lack of opportunities for women in AI and machine learning.
SoftBank's Masayoshi Son is seeking $100 billion to build a new venture in the AI chips industry, to compete with Nvidia.
The new venture, code-named Izanagi, would collaborate with Arm, the chip design company that SoftBank spun out last year.
SoftBank plans to tap Middle East-based institutional investors for $70 billion of the $100 billion, with SoftBank itself providing the remaining $30 billion.
Dili, a platform that uses AI, aims to automate key investment due diligence and portfolio management steps for private equity and venture capital firms.
The platform leverages AI models, such as large language models similar to ChatGPT, to streamline investor workflows and automate tasks like parsing databases and handling due diligence request lists.
Dili raised $3.6 million in venture funding and plans to expand into new applications, becoming an "end-to-end" solution for investor due diligence and portfolio management.
CapitalG, Alphabet's growth stage venture arm, has around 50 people on its team, with a large number of senior advisors from Alphabet who collaborate with its portfolio companies on technical and business matters.
CapitalG typically invests between $50 million and $200 million in each company and aims to be a long-term partner, focusing on market differentiation and scalability.
The firm is enthusiastic about AI and looks for companies with technical differentiation in areas where existing distribution is less important. It sees AI as a tool for enhancing the customer experience and rethinking marketing, customer support, and internal processes.
OpenAI has developed Sora, an AI app that can generate photorealistic videos based on text prompts, showcasing impressive photorealism and the ability to produce longer clips up to one minute in length.
Sora demonstrates an emergent grasp of cinematic grammar, producing videos with multiple shot changes, showing its ability to tell a story through camera angles and timing.
While Sora has restrictions on content and OpenAI plans to ensure safety, the app's potential to generate deepfakes and infringe on copyrighted work raises concerns that must be addressed.
Google inaugurated a new AI hub in Paris, which will house around 300 researchers and engineers from Google Research, DeepMind, YouTube, and Chrome.
The hub is part of Google's effort to attract AI talent and solidify its position as a leading AI company.
The move highlights Google's insecurity about AI and its desire to showcase its commitment to artificial intelligence amid stiff competition from other tech giants.
Kong has launched an open-source AI Gateway that allows developers and operations teams to integrate their applications with large language models (LLMs) through a single API. The gateway supports various LLM providers and includes AI-specific features, such as prompt engineering and credential management. Kong aims to make building applications with AI easier by providing a central point for managing guidelines, prompts, and API usage.
The AI Gateway enables developers to change prompts and results on the fly, making it easier to translate or remove personally identifiable information. Kong's API gateway allows for the consumption of multiple LLM providers without requiring changes to code. The company plans to introduce premium features in the future, but for now, the new AI features are available for free.
Bret Taylor, a board member of OpenAI, has founded a new startup called Sierra that specializes in building conversational AI agents.
FlowFi, a startup focused on helping startups manage their finances, is taking a counter-cultural approach by pairing its software with a labor marketplace to blend human and computer intelligence.
Bold and Antithesis, two startups in the fintech and software testing sectors respectively, have raised significant amounts of capital for their businesses.
OpenAI has partnered with Microsoft Threat Intelligence to disrupt five state-affiliated actors who were using AI services for malicious cyber activities.
The identified actors, which include groups from China, Iran, North Korea, and Russia, were using OpenAI services for various purposes such as researching companies, translating documents, and creating content for phishing campaigns.
OpenAI is taking a multi-pronged approach to combatting these malicious actors, including monitoring and disrupting their activities, collaborating with industry partners to exchange information, iterating on safety measures, and promoting public transparency.
TikTok is planning to launch localized election resources in its app for each European Union (EU) Member State to combat disinformation risks related to regional elections. The Election Centers will provide users with trusted and authoritative information and will be labeled to direct people to the relevant center. TikTok will also add reminders to hashtags to encourage users to follow guidelines, verify facts, and report content that violates community guidelines.
TikTok is expanding its media literacy campaigns in the EU and seeking to expand its fact-checking partner network. It currently works with nine organizations covering 18 languages, but efforts to address election security risks related to AI-generated deepfakes were not mentioned in the announcement.
ChatGPT is being tested with a new memory feature that allows it to remember information discussed in previous chats, making future conversations more helpful.
Users have full control over ChatGPT's memory and can choose to remember, ask, or forget information. They can also disable the memory feature completely if desired.
The memory feature is being rolled out to a small number of ChatGPT free and Plus users for testing, with plans for a wider release to be announced soon.
YC-backed startup Cambio is using AI bots to negotiate debt collections and to assist banks and credit unions with sales calls.
Cambio's AI-powered service has helped 70% of customers resolve their collections and raise their credit score.
The company's AI bots listen in on calls and provide real-time coaching on what customers should say to collectors in order to negotiate down their debt.
Former Salesforce co-CEO Bret Taylor and Google employee Clay Bavor have launched a conversational AI company called Sierra, which aims to go beyond traditional customer service bots.
Sierra's software can take actions on behalf of the customer, such as upgrading subscriptions and managing complex tasks like furniture deliveries.
The company claims to solve issues like hallucinations, where language models may provide incorrect answers, and is already working with brands like SiriusXM, Sonos, and WeightWatchers.
British sculptor Nick Hornby creates sculptures that transform depending on the viewer's perspective, raising questions about power and the role of monuments.
Hornby's sculptures are created using computer modeling and digital processes, with his equestrian sculpture consisting of 165 manipulated metal components.
The artist is now shifting focus to include himself in his work and plans to experiment with new technologies such as generative AI in future projects.
Bugcrowd, a bug bounty platform, has raised $102 million in an equity round led by General Catalyst. The startup connects organizations with over 500,000 hackers to identify bugs and vulnerabilities in their code.
The funding will be used to expand Bugcrowd's operations in the U.S. and internationally, potentially through M&A, and to enhance its platform with additional functionality.
Bugcrowd has been growing at over 40% annually, has more than 1,000 customers, and is approaching $100 million in annual revenues.
Venture capital firm Peak XV is taking portfolio companies on a trip to Silicon Valley to meet industry leaders and visit AI research centers. The week-long trip, called "Immersion Week," includes strategy sessions with OpenAI and Nvidia executives, as well as talks from experienced operators. Peak XV is broadening its offerings beyond just funding as it seeks access to promising AI startups globally.
The trip is particularly beneficial for startups in India, where there is a lack of depth in deeptech and AI startups. Many Indian startups are focusing on building new capabilities and finding customers overseas, making a trip to Silicon Valley valuable.
Peak XV's Surge batch consists of 77% AI and deep tech startups, highlighting its focus on these areas. The venture firm, with $2.5 billion to deploy in the region, has been aggressively building its bench strength and networking capabilities since its split from Sequoia Capital.
Stockholm-based startup Xensam has raised $40 million in its first round of funding to further develop its AI-powered software asset management tools. The funding will be used to expand the company's AI technology stack, hire more employees, and enter the US market. Xensam's approach involves using AI to comprehensively scan an organization's network and provide real-time insights into software usage across cloud and on-premise environments.
The software asset management market is experiencing rapid growth due to the increasing use of cloud computing and software-as-a-service. Xensam's AI technology sets it apart from competitors by organizing and normalizing large amounts of data, which can help companies identify overpayments, security vulnerabilities, and operational glitches.
Bootstrapped for eight years, Xensam's founders decided to seek external funding to support the growth of their business while still maintaining their cultural values. Expedition Growth Capital, a London-based investor, provided the funding and will work closely with the company to support its growth.
Precision-manufacturing startup Daedalus has raised $21 million in a Series A funding round led by Nokia-funded NGP Capital. The German company uses artificial intelligence to automate manual tasks and optimize workflows in the production of bespoke precision parts for industries such as medical devices, aerospace, defense, and semiconductors. Daedalus plans to open additional factories in Germany and expand internationally based on demand.
Founder and CEO Jonas Schneider, a former OpenAI engineering lead, aims to redefine manufacturing by applying machine learning to create high-end, industrial-grade parts that are not feasible with traditional 3D printing techniques.
Singapore-based startup Brilliant Labs has unveiled Frame, a pair of lightweight augmented reality (AR) glasses powered by its multimodal AI assistant, Noa. The glasses have received investment from John Hanke, CEO of Niantic, the company behind Pokémon GO. Frame's lenses have a resolution of 640 x 400 and can display videos and photos, while Noa is capable of visual processing, image generation, translation, and answering questions. Preorders for Frame are available now, and the glasses will retail for $349 and begin shipping in April.
Attentive.ai, a startup focused on landscaping and construction services, has raised $7 million in funding to enhance its AI-driven offerings and expand to more businesses.
The startup provides an end-to-end business management platform with AI-based workflows, allowing companies to save time and bid for outdoor contracts using automated site measurements.
Attentive.ai plans to expand its focus to construction operations and target general and subcontractors and suppliers in a tool called Beam AI, which delivers multiple construction estimates simultaneously by automating blueprint tracing.
Entrust is acquiring identity verification startup Onfido for a price rumored to be over $400 million. The deal is still going through regulatory approvals and the completion date is unclear. The plan is to integrate Onfido's AI-based tools into Entrust's technology stack, allowing them to have a leadership position in the identity verification market.
Entrust currently has nearly $1 billion in annual revenue with 10,000 customers worldwide, including governments and major banks. Onfido, founded in 2012, raised $100 million in funding in 2020 and saw increased demand during the Covid-19 pandemic as digital transactions and the need for digital identity verification grew.
The UK government plans to invest over $125 million in funding to boost AI regulation and innovation.
£10 million will be dedicated to helping regulators upskill in applying existing rules to AI developments and enforcing laws on AI apps.
An additional £90 million will be used to establish nine research hubs to foster homegrown AI innovation in areas such as healthcare and chemistry.
New research from MIT shows that encoding symmetries in machine learning models can help the models learn with fewer data.
The researchers modified Weyl's law to factor in symmetry when assessing the complexity of a dataset, leading to a reduction in the amount of data needed for learning.
Symmetries can provide linear improvements in sample complexity, but they can also yield exponential gains, especially in higher-dimensional spaces.
Cloud infrastructure market experienced significant revenue growth in Q4 2023, driven by interest in generative AI and technologies like ChatGPT. The cloud infrastructure market for the entire year reached $270 billion, up from $212 billion in 2022. Microsoft's investment/partnership with OpenAI is giving it an edge in the market, with a 2% increase in market share in Q4.
Researchers from Cornell University have analyzed the use of the autopen, a device used to automate signatures, and its impact on communication. They found that while the autopen made communication faster, it also instilled mistrust and reduced the perceived value of signed items. The researchers draw parallels to current concerns about AI technology, particularly in relation to the use of ChatGPT for communication.
Google has launched ImageFX, an AI-powered image generator that allows users to create and edit images using text prompts and "expressive chips."
Google has implemented safeguards to prevent the misuse of ImageFX, including limiting outputs of violent, offensive, and sexually explicit content and tagging images with a digital watermark for identification.
Imagen 2, the AI image model developed by Google's DeepMind team, is being expanded to more of Google's products and services, including its AI search experience and managed AI services.
Google has released MusicFX, an upgraded version of its music-generating tool, MusicLM. MusicFX can generate ditties up to 70 seconds in length, with "higher-quality" and "faster" results. The tool allows users to enter text prompts and provides alternative descriptors and recommendations for relevant descriptions and instruments.
Google has also released TextFX, a tool designed to aid in lyric writing. It includes modules that find words in a category starting with a chosen letter and modules that find similarities between two unrelated things. Google warns that TextFX may display inaccurate information.
Questions remain surrounding the use of AI-generated music and whether it violates copyright. Music labels have flagged AI-generated tracks to streaming partners, and there is still a lack of clarity on the legal implications of "deepfake" music. Google is trying to navigate this landscape in its deployment of GenAI music tools.
OpenAI is allowing users of ChatGPT to bring GPTs into conversations by typing "@". Users can select a relevant GPT from a list and add it to the conversation with full context.
The GPT Store, a marketplace for GPTs, was recently launched by OpenAI. Developers can create GPTs for various purposes, such as trail recommendations and code tutoring.
OpenAI is facing challenges with moderation, as some GPTs violate their terms by being sexually suggestive or impersonating individuals. The company is working on a combination of human and automated review to address these issues.
Startup Cap VC is launching an AI-powered tool for VC firms, with plans to expand to startups raising money. The tool aims to make investment decisions faster and more efficient by turning unstructured data from PDF files into structured data. Cap VC is also building a fund management tool for LPs and auditors to utilize.
The platform aims to provide VCs with a full context of their portfolio companies and the companies they might invest in, including historical data. Cap VC is leveraging insights from auditing firms like Deloitte to create a more accessible space for different stakeholders, including regulatory bodies.
The CEO of Cap VC believes that many VCs haven't built similar platforms themselves due to a lack of understanding the tech startup ecosystem and potentially being lazy. The startup aims to launch its platform to the public in February.
MLCommons has launched a new working group, MLPerf Client, with the goal of establishing AI benchmarks for desktops, laptops, and workstations running various operating systems.
The first benchmark will focus on text-generating models, specifically Meta's Llama 2, which has been optimized for Windows devices.
Members of the MLPerf Client working group include AMD, Arm, Asus, Dell, Intel, Lenovo, Microsoft, Nvidia, and Qualcomm, but not Apple.
Etsy has launched a new AI-powered feature called "Gift Mode" to help users find tailored gift ideas based on specific preferences. Users can take an online quiz that asks about the recipient's interests and generates gift guides inspired by their choices. The feature uses a combination of machine learning, human curation, and OpenAI's GPT-4.
The new feature aims to relieve the stress of selecting the perfect present, with 71% of respondents in Etsy's latest survey saying they felt anxious when shopping for gifts. Etsy plans to enhance Gift Mode's capabilities over time and become the destination for gifting.
Etsy has previously released gift-related offerings such as wedding and baby registries. The company is also investing in the gifting space and recently introduced a program called "Share & Save" to lower transaction fees for sellers.
TextQL is a platform that connects a company's existing data stack to large language models like OpenAI's ChatGPT and GPT-4, allowing business teams to ask questions of their data on-demand.
TextQL uses a data model to map a company's database to the "nouns" representing a customer's business, enabling users to ask questions and take actions based on the data.
The platform has gained traction in healthcare, bio and life sciences, financial services, manufacturing, and media industries, with several years of runway and annual recurring revenue in the six figures.
Artisse has raised $6.7 million in seed funding for its AI photography app that generates realistic photos of users using prompts or uploaded selfies.
The app has become popular due to the hyper-realistic images it produces, reaching an estimated 43 million people on social media and being downloaded over 200,000 times to date.
Artisse has plans to expand its AI technology to other areas, such as virtual fitting room tech for online shopping and a group photo feature that allows users to "pose" with celebrities.
A free tool called Kin.art has been launched to help prevent AI models from training on artwork without the artists' permission. The tool modifies the pixels of an image or conceals parts of the artwork to trick the AI models. Unlike other tools that mitigate the damage after the fact, Kin.art prevents artists' artwork from being included in AI training datasets in the first place.
Kin.art plans to offer the tool as a service in the future, allowing any website or platform to protect their data from unlicensed use. This philanthropic effort aims to help platforms that need to provide public-facing services and don't have the luxury of blocking non-users from accessing their data.
Chinese startup 01.AI has released an open-source AI model called Yi-34B that outperforms competitors on various language AI benchmarks and leaderboards. The startup aims to create the first "killer apps" built on the capabilities of language models and hopes to inspire a loyal developer base. The founder and CEO, Kai-Fu Lee, believes that the next-generation productivity tools should not resemble traditional office applications like Word, Excel, and PowerPoint.
OpenAI has signed its first higher education customer, Arizona State University (ASU), to bring its AI-powered chatbot, ChatGPT, to the university's researchers, staff, and faculty.
ASU will run an open challenge for faculty and staff to submit ideas on how to use ChatGPT, focusing on student success, research, and streamlining processes.
ASU will provide ChatGPT Enterprise accounts to its full-time employees, offering enhanced privacy and data analysis capabilities, admin tools, shareable conversation templates, and priority access to ChatGPT and Advanced Data Analysis.
Microsoft has made its AI-powered reading tutor, Reading Coach, free for anyone with a Microsoft account. The tool provides personalized reading practice and will soon integrate with learning management systems. Reading Coach allows learners to identify words they struggle with and provides tools for independent practice, including text-to-speech and picture dictionaries.
NASA has developed a self-assembling robotic structure for construction in space and on other planets. The structure uses cuboctahedral frames, called voxels, and two types of robots to assemble them. The robots, which can be charged wirelessly, can quickly and autonomously build structures of various angles and strengths.
The self-building structure has applications for lunar surface construction, including communication towers and shelters. It also has potential for long-duration or large-scale infrastructure projects in space and on other celestial bodies.
The robots developed by NASA took 4.2 days to assemble 256 voxels into a passable shelter structure. They could build larger structures with sufficient time, or affix plating to the exterior for additional functionality.
The European Parliament is calling for new rules to bring more fairness and transparency to music-streaming platforms.
The proposed bill would require streaming platforms to open up their recommendation algorithms and clearly indicate when a song has been generated by AI.
The aim is to ensure that European artists have more visibility and prominence on music-streaming platforms and to prevent manipulation of streaming figures that can impact artists' fees.
Google has introduced an AI-powered addition to its visual search capabilities in Google Lens, allowing users to ask questions about what they see and receive generative AI-powered answers. The feature offers insights and information based on web searches and can be activated through a gesture called Circle to Search. While the feature aims to improve search results, the accuracy and relevancy of the answers may not always be guaranteed.
OpenAI is forming a Collective Alignment team to collect and incorporate public input on its AI models' behaviors into its products and services.
The team is an extension of OpenAI's public program, launched last year, which awarded grants for experiments in establishing a democratic process to decide rules for AI systems.
The code used in the grant recipients' work has been made public, along with summaries of each proposal and takeaways.
This podcast episode explores the ethical challenges of autonomous vehicles.
The transformative impact of AI in sectors like healthcare and finance is discussed.
The episode offers a wide range of insights for listeners interested in AI.
Google is committed to protecting the integrity of elections by preventing abuse and ensuring the responsible use of their AI tools.
They are investing in initiatives to prevent misleading "deepfakes", influence operations, and chatbots impersonating candidates.
Google is also focusing on transparency by improving image provenance and integrating ChatGPT with reliable news sources, while providing access to authoritative voting information through partnerships with organizations like the National Association of Secretaries of State.
Microsoft has launched a consumer-focused paid plan for its AI-powered content-generation technology, Copilot, as well as expanding eligibility for enterprise-level offerings.
The new consumer plan, called Copilot Pro, is priced at $20 per user per month and grants access to Copilot GenAI features across Microsoft 365 applications, including Word, Excel, PowerPoint, Outlook, and OneNote.
Copilot Pro subscribers also receive 100 "boosts" per day in Designer, Microsoft's AI-powered image creation tool, and have priority access to the newest GenAI models underpinning Copilot.
Google is laying off over 1,000 employees, including those in the voice-activated Google Assistant and hardware teams.
The company is restructuring its knowledge and information product team, as well as the Devices and Services PA (DSPA) team responsible for managing Pixel, Nest, and Fitbit hardware.
Fitbit co-founders James Park and Eric Friedman are leaving as part of the restructuring.
OpenAI has launched the GPT Store, an app store where users can create and publish their own custom versions of ChatGPT, adding functionality to the chatbot.
The GPT Store is only accessible to users with a ChatGPT Plus subscription or the business plans ChatGPT Team and Enterprise.
OpenAI has yet to reveal how app makers will be paid, but developers are taking a leap of faith due to the popularity and hype of ChatGPT.
OpenAI has launched ChatGPT Team, a self-serve plan that provides access to advanced models like GPT-4 and DALL·E 3, along with tools like Advanced Data Analysis. It offers a dedicated workspace for team collaboration and administration tools for team management.
Users of ChatGPT Team have the ability to customize ChatGPT for specific tasks, such as project management, onboarding, data analysis, and more. It allows teams to improve efficiency and work quality, as shown by a study where employees reported completing tasks faster and achieving higher quality work with access to GPT-4.
ChatGPT Team is priced at $25/month per user when billed annually or $30/month per user when billed monthly, and users can upgrade in their ChatGPT settings to get started.
Investors are becoming more cautious about investing in AI startups and are looking for companies that will turn a profit.
Building a profitable AI business comes with challenges such as high costs, talent shortages, and expensive API and hosting requirements.
To build a profitable AI startup, it is important to have a realistic cost model, determine whether to use a cloud-based AI model or host your own, and consider the long-term financial viability of your business model.
Twitter (referred to as "X" in the article) has a problem with verified bots, despite the suggestion that forcing users to pay for verification would solve the issue. A video on Instagram Threads highlights numerous bots, including verified ones, posting an AI-generated response that goes against OpenAI's policies.
There are suspicions that some of the bot activity could be coming from Twitter itself, as older, abandoned accounts are being turned into verified bots with AI automation.
Twitter's bot problem extends beyond AI-powered accounts, with many bots operating without OpenAI's assistance and being harder to detect. The company previously admitted to having a Verified spammer problem and introduced new DM settings to address the issue.
Getty Images launches Generative AI by iStock, a service that uses AI models trained on Getty's iStock stock photography and video libraries to generate new licensable images and artwork, while preventing copyright infringement. The service can modify existing images and generate new ones in 75 languages and offers integration with existing apps. It costs $15 for every 100 generated images and comes with $10,000 in legal coverage for any licensed visuals generated.
The launch of Generative AI by iStock comes amid a growing debate around copyright infringement by AI-generated content. Some companies argue fair use doctrine protects them, but lawsuits against companies like Stability AI and Getty Images suggest the issue is far from settled.
Vendors are starting to offer legal fee coverage to customers facing copyright lawsuits from using GenAI tools, and Generative AI by iStock offers $10,000 in legal coverage to its customers.
Walmart debuts two generative AI-powered tools: a search feature that allows customers to search for products by use cases instead of product names, and an AI-powered replenishment tool that creates automated shopping carts for items customers regularly order.
The company also introduces "Shop with Friends," an AR social commerce platform that allows customers to share virtual outfits they create and get feedback from friends.
Walmart highlights the use of AI in other areas of its business, including Sam's Club, where AI and computer vision are used to speed up receipt verification, and in-store associates who can use an AI tool called My Assistant for writing and summarizing documents.
Microsoft has collaborated with the Pacific Northwest National Laboratory to utilize its Azure Quantum Elements service to narrow down potential new battery materials, resulting in one prototype. Azure Quantum Elements combines AI and traditional high-performance computing techniques for scientific computing. Although no quantum computer was used in this project, the overall goal is to bring these technologies together in the future.
Using Azure Quantum Elements, the researchers were able to go through the process of identifying battery candidates and building a prototype in just 18 months, a process that typically takes years.
Microsoft remains optimistic about building a quantum supercomputer within the next decade, but currently, quantum computing is still far from being integrated into scientific processes.
Samsung announces new features for its SmartThings home automation platform, including a dashboard screen called Now Plus that displays information about connected devices and the indoor temperature on Samsung TVs.
The SmartThings platform now includes a "map view" feature that shows an interactive map of the home with the location of smart devices, and it can be created manually or with a photo of an existing floor plan.
The map view also includes "AI characters" that represent family members and pets, and these characters respond to real-time conditions in the home.
Isomorphic Labs, a spin-out of DeepMind, has partnered with pharmaceutical companies Eli Lilly and Novartis to use AI for drug discovery. The deals are worth a combined $3 billion, with Isomorphic receiving $45 million upfront from Eli Lilly and potentially up to $1.7 billion based on performance milestones. Isomorphic will utilize DeepMind's AlphaFold 2 AI technology to predict the structure of proteins and identify new target pathways for drug treatment.
The partnerships aim to transform the discovery of new drugs and accelerate the development of life-changing medicines. Isomorphic's collaboration with Eli Lilly and Novartis combines AI and data science with medicinal chemistry and disease area expertise.
The latest version of AlphaFold can generate predictions for almost all molecules in the Protein Data Bank and accurately predict the structures of ligands, nucleic acids, and post-translational modifications. Isomorphic is already applying this new model to therapeutic drug design.
Google's DeepMind Robotics researchers are exploring new methods to give robots a better understanding of human intentions using large language models and video input.
The new AutoRT system uses large foundational models to provide better situational awareness to robots, allowing them to manage a fleet of robots working together and understand natural language commands.
The RT-Trajectory system leverages video input and overlays a two-dimension sketch of the arm in action, providing visual hints for the robot as it learns its control policies.
AI-powered search engine Perplexity AI has raised $70 million in funding, valuing the company at $520 million.
Perplexity offers a chatbot-like interface where users can ask questions in natural language and receive AI-generated summaries with source citations.
The search engine startup aims to offer robust search filtering and discovery options, and is also developing its own gen AI models through an API for improved performance.
Intel is spinning out a new enterprise-focused gen AI software company called Articul8 AI, in partnership with DigitalBridge. The platform builds off a proof-of-concept from an Intel collaboration with Boston Consulting Group (BSG) and is optimized for speed, scalability, security, and sustainability. Arun Subramaniyan, formerly of Intel, will be the CEO of Articul8.
Sales professionals need to use AI to win the tug of war between sellers and buyers in the sales pitch.
AI can analyze data and provide insights to help sales professionals tailor their pitches to individual buyers.
AI can also automate time-consuming tasks, allowing sales professionals to focus on building relationships with buyers.
GitHub has made its programming-centric chatbot, Copilot Chat, generally available to all users. The chatbot is powered by OpenAI's GPT-4 model and can provide real-time guidance for developers, explain concepts, detect vulnerabilities, and write unit tests.
Codebase owners cannot opt out of their code being used for training the AI model, but GitHub suggests making repositories private to prevent inclusion in future training sets.
GitHub's competitor, Amazon, offers a similar tool called CodeWhisperer, which has been upgraded with enhanced suggestions for app development on MongoDB and offers free usage to developers.
China's robotaxi startups, including Deeproute.ai, WeRide.ai, Pony.ai, and Momenta, are shifting their focus from full self-driving technologies to more commercially viable smart-driving solutions as monetization becomes urgent and the prospect of going public in the US becomes uncertain. The widespread availability of robotaxis remains a distant reality due to challenges such as safety, regulations, and costs. These companies are now seeking alternative revenue streams, such as selling advanced driver assistance systems (ADAS) to automakers and forming partnerships with original equipment manufacturers (OEMs) or relying on government contracts.
While full self-driving technology promises potential billion-dollar businesses, the revenues from selling to OEMs with ADAS are limited. The market for ADAS is smaller, and OEMs are less keen to work with software companies as they develop their own solutions. Building partnerships with OEMs is a lengthy and complex process that requires significant customization and buy-in from various stakeholders within the OEM. Some companies are also depending on government contracts for survival.
In addition to China, some robotaxi startups, such as Pony and WeRide, are exploring overseas markets, particularly in the Middle East, which is seen as an untapped market with friendly regulations and funding opportunities. However, the success
Generative AI companies are facing similar challenges to social media platforms, including issues with content moderation, labor practices, and disinformation. They are built on problematic infrastructures and often rely on outsourced workers with low pay and difficult working conditions.
AI companies are struggling to effectively respond to criticism and address the unintended consequences of their technology. Policies and safeguards are easily circumvented, and measures such as digital watermarks are unlikely to be long-term solutions.
Generative AI has the potential to exacerbate the spread of misinformation and deepfakes, making it faster, cheaper, and easier to produce false content. This undermines the veracity of real media and information, and the problem is further amplified by the reduction in resources and teams dedicated to detecting harmful content.
Chinese AI startup DP Technology is focused on applying AI to molecular simulations, believing that "scientific research for humanity" will drive its global expansion. DP provides tools for scientific computing, combining machine learning with molecular simulations to solve problems in the physical world. It plans to expand to the US market, starting with opening an office and working with a partner to distribute its products and services.
Researchers from MIT have used deep learning to discover a class of compounds that can kill drug-resistant bacteria, such as MRSA. The compounds were shown to have low toxicity against human cells, making them promising drug candidates. The researchers were able to identify the chemical structures of the compounds and gain insights into how the deep learning model made its predictions, which could aid in the design of new antibiotics.
VERSES AI makes a breakthrough in developing AGI based on 'natural' intelligence rather than 'artificial' intelligence.
VERSES AI appeals to Open AI to collaborate on building AGI in a safe and beneficial manner for humanity.
Open AI's charter includes a commitment to assist a value-aligned, safety-conscious project if it comes close to building AGI before them.
Researchers have developed computational event-driven vision sensors that can convert motion into spiking signals. These sensors combine event-based sensing with spiking neural networks, enabling them to perform both sensing and computation tasks without the need for data transfer. The sensors reduce data redundancy, improve energy-efficiency, and enable real-time information processing for applications like autonomous driving and intelligent robotics.
MIT researchers have developed a new metric called "minimum viewing time" (MVT) to quantify the difficulty of recognizing images. The MVT metric measures the minimum amount of time a person needs to view an image before they can accurately identify it. Existing benchmark datasets, including those specifically designed to challenge AI models, tend to be skewed towards easier images, which may not reflect real-world performance.
Agility is using large language models to improve communication and programming for its bipedal robot, Digit.
The company has developed a demo that allows Digit to understand natural language commands of varying complexity and execute tasks accordingly.
This use of generative AI and large language models has the potential to make robots more versatile and faster to deploy in real-world scenarios.
OpenAI's Superalignment team is working on ways to control superintelligent AI systems that have intelligence beyond that of humans.
The team is focused on developing governance and control frameworks to steer AI systems in desirable directions and away from undesirable ones.
OpenAI is launching a $10 million grant program to support research on superintelligent alignment and will make its research and code publicly available.
Spotify is testing a new feature that allows users to create playlists using AI technology and prompts. The AI playlists feature can be accessed through the "Your Library" tab on the Spotify app, giving users the option to type their own prompt or choose from suggested prompts. Spotify has been investing in AI across its app, including personalized playlists and an AI DJ, but has not confirmed a launch date for the AI playlists feature.
Google introduces MedLM, a family of generative AI models specifically designed for the medical industry.
MedLM includes a larger model for complex tasks and a smaller, fine-tunable model for scaling across tasks.
Early users of MedLM, such as HCA Healthcare and BenchSci, have found success in using the models to aid healthcare professionals in tasks like drafting patient notes and identifying novel biomarkers.
OpenAI has released GPTs, which are customized versions of ChatGPT designed for specific purposes. Building a GPT does not require coding skills and can be done through a simple conversation and the selection of capabilities.
Users can initiate a conversation, provide instructions and additional knowledge, and customize a GPT to meet their specific needs.
GPTs are versatile and can be used for a wide range of applications, from drafting emails to creating code tutorials.
New EU rules will prevent platforms, such as Uber and Deliveroo, from automatically firing their workers, improving labor rights for millions of gig economy workers in Europe.
The rules will clarify whether platform workers should be considered employees or independent contractors and grant them social rights, such as sick pay and holiday pay.
Platform workers will be legally considered as employees if their relationship with the platform meets two out of five criteria, including task allocation and performance supervision.
Researchers from MIT and the Chinese University of Hong Kong have developed a machine learning technique called neural lithography that uses a digital simulator to close the gap between the design and manufacturing of optical devices. The simulator, which incorporates real data gathered from a photolithography system, accurately models the specific deviations of the system and allows for the production of devices that better match their design specifications. This method has the potential to improve the accuracy and efficiency of optical devices used in applications like mobile cameras, augmented reality, and medical imaging.
MIT professor Justin Solomon applies modern geometric techniques to solve problems in computer vision, machine learning, statistics, and beyond.
Solomon's research group works on problems that involve processing geometric data and high-dimensional statistical research using geometric tools.
Solomon is passionate about making geometric research accessible to underserved students and has launched the Summer Geometry Initiative to provide research opportunities.
Spore.Bio, a French startup, has developed a pathogen detection methodology that uses deep-learning algorithms to speed up the process of detecting contamination in food factories.
The technology works by shining light on surfaces where clean and unclean food has been and comparing the two data sets to identify when a surface is contaminated.
The startup has raised €8 million in pre-seed funding and plans to use the funding to develop a handheld device that can detect pathogens in real-time on the factory floor.
China's autonomous vehicle company WeRide has obtained two licenses from Singapore, allowing its robobuses to test on public roads in certain areas. The licenses will enable WeRide to test its self-driving buses in areas including the One North tech cluster and the National University of Singapore. WeRide has been strategically expanding globally and has secured licenses and permits in countries such as the United Arab Emirates, the U.S., and China.
Singapore is gearing up to enter the second phase of autonomous vehicle development, allowing AVs for passenger and utility purposes to operate in selected areas. WeRide's licenses align with Singapore's measured approach to rolling out AVs and its focus on creating controlled environments for testing. The country has attracted other global players in the autonomous vehicle industry, such as the Aptiv-Hyundai joint venture Motional.
WeRide has been actively building relationships with regulators and business partners in Singapore as part of its expansion strategy. The company has garnered significant investment from major public transport operators and local investment firms in Singapore. WeRide's expansion into Singapore is seen as a key step in its Asia-Pacific market expansion.
EU lawmakers have reached a political deal on a risk-based framework for regulating artificial intelligence (AI), which will result in a pan-EU AI law being implemented. The agreement includes a total prohibition on the use of AI for certain purposes such as biometric categorization systems using sensitive characteristics and social scoring based on personal characteristics. The law also includes obligations for AI systems classified as "high risk" and penalties for non-compliance.
The deal allows for a phased entry into force, with different requirements coming into effect at different times. The law is expected to come into full force in 2026. The EU's internal market commissioner described the agreement as "historic" and the first international regulation for AI in the world.
The UK Competition and Markets Authority (CMA) has launched an inquiry into the relationship between Microsoft and OpenAI and whether the two companies are in a "relevant merger situation". This inquiry comes after Microsoft made a significant investment in OpenAI, giving it nearly 50% ownership, and the two companies work closely together in the development of AI services. The CMA is concerned about the impact of this relationship on competition in the market and has opened an "Invitation to Comment" to gather feedback from the companies and interested third parties.
Computer scientists at Johns Hopkins University have developed a new approach to reducing the size of multilingual language models (MLMs) without compromising their performance. Their method, called Language-Specific Matrix Synthesis, reduces the number of parameters required for an MLM to function in each new language by using low-rank matrices. This allows for the deployment of smaller language models capable of handling hundreds of languages on a single device.
The new method achieves superior performance in multilingual settings while using fewer parameters, resulting in a significant reduction in a language model's size.
The reduced hardware requirements of a smaller language model make it feasible to deploy truly multilingual AI models on devices of all sizes, according to the researchers.
Google's Gemini AI model is now available in the Bard chatbot, allowing users to try it for free.
Gemini AI is currently only available in English but will support other languages soon.
Future releases of Gemini will include multimodal capabilities, processing multiple types of input and producing different outputs.
Meta has launched a standalone AI-powered image generator called Imagine with Meta. Users can create high-resolution images by describing them in natural language, similar to OpenAI’s DALL-E. Meta plans to add invisible watermarks to the generated content for increased transparency and traceability.
Meta has fully rolled out its AI characters across its U.S. apps, allowing users to chat with characters based on real-life celebrities such as Paris Hilton, Mr. Beast, and Kendall Jenner. The AI characters will now have long-term memory, meaning they can remember previous conversations, and will also support Bing Search. Users can access the AI characters by starting a new message and selecting "Create an AI chat" on Instagram, Messenger, and WhatsApp.
The addition of long-term memory gives the AI characters a more realistic feel and allows Meta to retain user data to improve its AI products over time.
More of Meta's AI characters will now support Bing Search, including characters based on Tom Brady, Charli D'Amelio, and Naomi Osaka. Users will be able to continue conversations with the AI characters from where they left off.
Other AI character apps, such as Character AI, founded by Google AI researchers, will provide competition for Meta's AI characters. Character AI recently raised $150 million in funding.
MIT spinoff Liquid AI has raised $37.5 million in seed funding to build general-purpose AI systems using liquid neural networks, a new type of AI model that is smaller and requires less compute power to run than traditional models. Liquid neural networks also have the unique ability to adapt their parameters over time, making them more effective in navigating changing environments. The startup plans to commercialize the architecture and provide on-premises and private AI infrastructure for customers to build their own models.
Google announces the launch of its new Cloud TPU v5p, the most powerful AI accelerator yet.
The v5p features a 2x improvement in FLOPS and 3x improvement in high-bandwidth memory compared to previous versions.
The TPU v5p can train large language models 2.8 times faster than the TPU v4 and is more cost-effective.
Google DeepMind has announced its new AI model, Gemini, which is described as a "multimodal" model, capable of processing information in various formats, such as text, audio, images, and video. The model has the potential to advance robotics and other AI projects.
Gemini is a significant step forward in AI models inspired by the way humans interact and understand the world through their senses. It combines different forms of data, allowing for complex reasoning and the integration of text, audio, images, and video.
Google has developed Gemini at a fast pace in response to competition from OpenAI. Gemini aims to surpass the capabilities of OpenAI's GPT-4 and bring AI models closer to real-world physical interaction, possibly leading to advancements in robotics.
Titan AI, a mobile games studio, has raised over $500,000 in pre-seed funding to develop generative AI technology that streamlines mobile game development, reducing cost and increasing speed.
The company uses Stable Diffusion and DALL-E image generators to create 2D graphics, combining them with proprietary technology to develop 3D models and level segments.
Titan AI aims to create more inclusive gaming experiences and has launched several game prototypes, including Aztec Spirit Run, which features a protagonist defending treasures against Conquistadors.
EnCharge AI has raised $22.6 million in a recent funding round to further develop its AI-accelerating chips and "full stack" AI solutions.
The startup aims to provide more affordable and energy-efficient AI chips to expand access to AI for organizations that cannot afford current costly and energy-intensive options.
EnCharge's hardware uses in-memory computing to accelerate AI applications in servers and network edge machines while reducing power consumption, but it has yet to mass produce its chips and faces competition from well-financed rivals in the AI accelerator hardware market.
JavaScript has found a significant role in the field of Artificial Intelligence (AI), showcasing its versatility and effectiveness in building intelligent systems.
JavaScript seamlessly integrates with web technologies, making it an ideal choice for AI applications that require web-based interfaces.
Embracing JavaScript in AI development opens up a world of possibilities as the boundaries between web development and AI continue to blur.
Mastercard has launched a new AI-powered shopping tool called "Shopping Muse" that provides personalized product recommendations based on users' colloquial language and shopping context.
The tool uses generative AI and algorithms that analyze data from the retailer's product catalog, the user's on-site behavior, and their past purchase and browsing history to provide tailored recommendations.
Shopping Muse's advanced image recognition tools also allow it to recommend relevant products based on visual similarities, even without precise technical tags.
OpenAI has signed a letter of intent to purchase $51 million worth of brain-inspired chips from Rain AI, a startup in which OpenAI CEO Sam Altman has personally invested. The chips, called neuromorphic processing units (NPUs), are designed to replicate features of the human brain and could potentially provide significantly more computing power and energy efficiency than traditional graphics chips (GPUs) used in AI development.
Rain AI has projected that it could release its first hardware to customers as early as October next year. However, the company has recently faced challenges, including a reshuffling of leadership and the forced removal of a Saudi Arabia-affiliated fund as a stakeholder. These challenges could potentially delay the delivery of the chips to OpenAI.
Altman has discussed raising money to start a new chip company in the Middle East, aiming to diversify OpenAI's chip sources beyond its reliance on GPUs and specialized chips from companies like Nvidia, Google, and Amazon.
Researchers have used artificial intelligence (AI) to develop a method for predicting optimal drug synthesis, potentially reducing the number of required lab experiments and increasing efficiency in chemical synthesis.
The AI model was trained on data from scientific works and experiments from an automated lab, and successfully predicted the position of a chemical transformation in drug molecules.
The method has been used to identify positions where additional active groups can be introduced in existing active ingredients, helping researchers develop new and more effective variants of known drugs.
Amazon CTO Werner Vogels predicts that generative AI will become culturally aware, understanding different cultural traditions and aspects. He believes that solving this issue is crucial for deploying this technology worldwide.
Vogels also believes that generative AI will greatly enhance developer productivity, providing tools that offer a broader view of things and serving as a senior developer who knows everything about a given code base.
In addition to AI, Vogels predicts that women's health tech will take off, driven by a societal shift and the increased flow of venture capital into this market. He sees a future where precision healthcare becomes the norm for women's healthcare.
Amazon has released an AI-powered image generator called Titan Image Generator, which can create new images based on text descriptions or customize existing images.
The generator can swap out backgrounds and generate lifestyle images while retaining the main subject.
It includes built-in mitigation for toxicity and bias and will protect customers accused of violating copyright with images generated by the tool.
Amazon AWS has launched SageMaker HyperPod, a purpose-built service for training and fine-tuning large language models (LLMs).
SageMaker HyperPod allows users to create distributed clusters that speed up the training process by efficiently distributing models and data across the cluster.
The service includes fail-safes to prevent the failure of the entire training process when a GPU goes down, and can speed up the training process by up to 40%.
Pinterest is testing a new tool that allows users to filter searches by different body types, as part of its efforts to make its platform more inclusive. The tool uses computer vision technology to identify various body types across images on Pinterest's platform. It aims to address body size discrimination and the negative impact of social media on body dissatisfaction, particularly among women and young people.
Pinterest's body type ranges tool is rolling out to users, starting with women's fashion and wedding ideas. The tool aims to improve search results' diversity and increase engagement on the platform. Pinterest's previous inclusive AI efforts, such as skin tone ranges and hair pattern search filters, have already shown positive results in terms of user engagement.
AR platform Squint has secured $13 million in a Series A funding round led by Sequoia and featuring Menlo Ventures. The platform connects users with detailed, step-by-step instructions when pointing their smartphones or tablets at physical objects, whether "smart" and connected or not. Squint is currently focused on business users and counts companies including Volvo and Siemens as clients.
California's Privacy Protection Agency (CPPA) has released draft regulations for the use of automated decision-making technology (ADMT) or AI. The regulations aim to give individuals control over their personal information used for automation and AI technologies. The proposed rules include opt-out rights, pre-use notice requirements, and access rights, which would allow California residents to understand how their data is being used for automation and AI.
The regulations could impact adtech giants like Meta, as businesses may be required to offer California residents the ability to opt out of their data being used for behavioral advertising. The CPPA's approach to regulating ADMT is risk-based and takes inspiration from the European Union's General Data Protection Regulation (GDPR).
The CPPA's regulations include transparency requirements, including pre-use notices and access rights, which allow individuals to access details about the use of ADMT and the logic behind automated decisions. The regulations also propose a threshold for decisions with legal or significant effects, profiling employees and students, and profiling in publicly accessible places, among other criteria.
AI startup PhysicsX has emerged from stealth mode with $32 million in funding. The London-based company has developed an AI platform that can create and run simulations for engineers working in industries such as automotive, aerospace, and materials science manufacturing. The funding will be used for business development and further development of the company's platform.
PhysicsX aims to solve a long-standing problem in manufacturing and physical production by using AI to simulate and test new ideas before they are developed. The platform allows engineers to predict the physics of a system with high accuracy and speed, enabling optimization and problem-solving in industries such as mining and engineering.
NVIDIA has launched a cloud service for medical imaging AI that provides APIs for developers and platform providers to integrate AI into their medical imaging offerings. The service includes pretrained AI models, annotation tools, and automated segmentation capabilities to streamline the development of medical imaging solutions. Solution providers and platform builders, such as Flywheel, RedBrick AI, and Dataiku, are already integrating the NVIDIA MONAI cloud APIs into their offerings.
French startup Osium AI has raised $2.6 million in seed funding to use artificial intelligence for research and development in materials science. The company aims to optimize the feedback loop between materials formulation and testing, helping industrial companies predict the physical properties of new materials and refine and optimize them. Osium AI is already in talks with 30 potential industrial clients.
OpenAI has announced a new "initial" board of directors that includes Bret Taylor as chair, Larry Summers, and Adam D'Angelo. The board is still subject to clarification and may change in the future.
Larry Summers, an economist and political veteran, brings valuable connections to governments, businesses, and academia to OpenAI.
Ilya Sutskever, OpenAI's chief scientist, is reportedly one of the key figures who pushed for former CEO Sam Altman's removal from the board and has experienced a loss of influence in the company.
Li Auto, a Chinese electric vehicle maker, is seeking talent in Singapore to develop automotive chips, specifically silicon carbide power modules. The company is hiring a general manager to establish its R&D center and formulate tech and product roadmaps for power semiconductors. Li Auto is among the young EV upstarts in China, along with Nio and Xpeng, and has shown strong sales figures with over 100,000 vehicles shipped in the third quarter.
Chinese EV firms are investing in making their own chips due to supply chain stability concerns and potential chip sanctions. Li Auto, Xpeng, and Nio are following in the footsteps of Tesla by developing their own chips. Wu, the former head of autonomous vehicles at Xpeng, has recently joined Nvidia, suggesting the semiconductor giant's focus on the auto chip market.
OpenAI's board is in discussions with Sam Altman, co-founder and ex-Y Combinator president, about returning as CEO.
Investors, including Thrive Capital and Sequoia Capital, are pushing for Altman's return to resolve the management crisis.
Altman has demanded "significant" changes in management and governance as a condition for returning.
Former OpenAI president Greg Brockman is announcing updates about OpenAI products, including the availability of ChatGPT's voice narration feature for all users.
The voice feature for ChatGPT is powered by a text-to-speech model and offers human-like voices generated from text.
Users can activate the voice feature by going to the settings menu in the Android or iOS ChatGPT apps.
Microsoft CEO Satya Nadella suggests that Sam Altman might return to OpenAI, despite Altman announcing his intention to join a newly-formed AI research team at Microsoft.
Nadella also expresses the desire for changes in governance at OpenAI, including around its investor relations, and states that Microsoft is open to both options of OpenAI employees staying at OpenAI or joining Microsoft.
The situation at OpenAI is unstable, with management and employees in revolt, and a board search for a new CEO resulting in a controversial choice. Over 700 employees have signed a letter calling for the board to resign and reinstate Altman.
OpenAI experienced a major shakeup recently, including the firing of Sam Altman and appointing Emmett Shear as interim CEO.
Sam Altman and others from OpenAI are heading to Microsoft, and Ilya Sutskever at OpenAI seems to be retracting everything that has happened.
The situation at OpenAI is still unfolding, and additional updates are expected in the coming days.
Headline, a VC fund, has developed an analytics tool called Deepdive to help founders determine whether they have achieved true product-market fit. Deepdive goes beyond revenue metrics and focuses on understanding customer behavior, retention dynamics, and spending patterns to provide a comprehensive view of a business's performance.
The goal of Deepdive is to shift the focus from pure revenue metrics to the value of each cohort, encouraging founders to prioritize product-market fit and responsible scaling. The tool is currently offered for free and Headline hopes to create a shared understanding of product-market fit in the startup ecosystem.
OpenAI has fired its CEO, Sam Altman, and appointed Mira Murati as interim CEO.
Murati has a background in mechanical engineering and previously worked at Tesla and Leap Motion before joining OpenAI as VP of applied AI and partnerships.
Murati believes that multimodal models, such as OpenAI's GPT-4 with Vision, are the future of the company and sees value in testing and understanding the limitations of AI technology.
1. Managing online reputation is crucial for businesses in the digital age, as it influences consumer decisions, builds confidence, and promotes loyalty.
2. Social media and online platforms have amplified the importance of reputation management in our digital lives.
3. The rise of AI technology offers opportunities to efficiently monitor and manage online reputation, making the market safer for consumers.
Venture firm Menlo Ventures has raised $1.35 billion in new capital for investments in AI startups.
The firm believes that AI will become a standard collaborator in daily tasks and expects the most exciting innovations to emerge in the AI field.
Menlo Ventures has a track record of successful investments, including early bets on companies like Uber and Warby Parker, and plans to continue investing in healthcare, consumer, cybersecurity, and fintech startups.
Blackshark.ai has developed Orca Huntr, an AI-powered tool that allows users to find and track objects from orbit using a simple interface. The tool allows users to label objects with a few brush strokes, refining the model in real-time to detect similar instances of the object. This no-code approach simplifies object detection and saves time compared to traditional labeling processes.
Blackshark.ai's Orca Huntr tool is built on the company's expertise in interpreting and using aerial and orbital imagery. It allows users to easily identify and track objects such as buildings with solar panels, burned areas in wildfire zones, and fishing boats in the ocean. The tool's simplicity and accuracy make it suitable for various applications, from real estate and development to military intelligence.
The company has secured $15 million in funding from investors including Point72 Ventures, M12 Microsoft's Venture Fund, and Maxar. With access to Maxar's archive of satellite data, Blackshark.ai can train models using a wide range of imagery, making it a valuable tool for governments and organizations that need to analyze geospatial data. The funding will help the company expand its capabilities and offer more comprehensive solutions to its clients.
Instagram introduces new features including custom AI stickers, photo filters, a clip hub, and more.
Users can now create custom stickers for Reels and Stories using AI, allowing them to "cut out" objects from their photos or videos.
Additional features include new photo filters, text-to-speech voices, text fonts and styles, access to trending audio, streamlined views of drafts, and improved Reels metrics.
Microsoft has revealed two custom-designed AI chips: the Azure Maia 100 AI Accelerator for training AI models and the Azure Cobalt 100 CPU for running them.
The chips, which will roll out early next year, are part of Microsoft's effort to optimize its datacenters for AI innovation and address the shortage of GPUs.
Microsoft's investment in custom chips aims to increase performance, power efficiency, and cost-effectiveness for customers using Azure services.
Microsoft has renamed Bing Chat, an AI-powered chatbot, to Copilot in Bing and the premium version to Copilot, to create a unified Copilot experience for consumers and commercial customers. Users signing into Bing with a corporate account will receive "commercial data protection," with their data not being saved or used to train AI models, and Microsoft won't have access to it. Copilot will now be accessible in Windows and available in Microsoft's enterprise subscription plans.
Vienna-based startup Atlas has emerged from stealth after two years with $6 million in seed funding to accelerate the development of its 3D generative AI platform. The platform partners with game developers and brands to build virtual worlds efficiently, allowing developers to generate detailed 3D models from reference images and text. Atlas aims to serve as a collaborative design partner and plans to target small and indie game developers with its technology.
YouTube is introducing new policies and tools to handle AI-created content on its platform, including requirements for creators to disclose when they have created altered or synthetic content that appears realistic. The company wants viewers to have context when viewing realistic content, especially when it discusses sensitive topics like elections. YouTube itself is using AI-generated content and will label all generative AI products and features as altered or synthetic.
OpenAI is partnering with outside organizations to create new data sets for training AI models in order to combat the flaws and biases in current data sets.
The Data Partnerships program aims to collect large-scale data sets that reflect human society and are not easily accessible online.
OpenAI is looking to create both open source and private data sets, with a focus on understanding different subjects, industries, cultures, and languages.
OpenAI is introducing OpenAI Data Partnerships, where they will collaborate with organizations to create public and private datasets for training AI models.
They are seeking large-scale datasets that reflect human society and are not easily accessible online. They can work with various types of data, including text, images, audio, and video.
There are two ways to partner with OpenAI: contributing to the creation of an open-source dataset for language models or providing private datasets to train proprietary AI models, with data privacy and access controls.
Bullet Points:
Picsart Ignite, a suite of AI-powered tools, has been launched by photo-editing startup Picsart. The suite includes 20 tools that make it easier to create ads, social posts, logos, and more. It also includes features such as AI Expand, AI Object Remove in Video, AI Style Transfer, and AI Avatar.
Picsart Ignite is now available to all users across web, iOS, and Android platforms.
Picsart has previously released AI-powered tools and recently launched in-app communities called "Spaces" for social collaboration.
AI startups in France are gaining momentum and attracting investor interest.
France has a strong talent pool of AI researchers and engineers, making it an ideal location for tech giants' AI research labs.
European AI startups are prioritizing regulation and compliance from the start, setting them apart from larger AI giants.
The Screen Actors Guild-American Federation of Television and Radio Artists (SAG-AFTRA) has reached a deal with the Alliance of Motion Picture and Television Producers to end the 118-day strike. The terms of the deal are unclear but it is expected to have long-lasting impact on the use of artificial intelligence in actors' performances and residual payments for streaming content.
This strike is the longest in Hollywood's history and has caused a significant disruption in the industry. The Writers Guild of America also went on strike earlier this year, marking the first time since 1960 that both writers and actors have faced a dual work stoppage.
Artificial intelligence was a major point of contention during the negotiations, with studios seeking to use AI scans of actors and performances without consent or proper compensation. It is likely that the actors have gained some protections regarding the use of AI in their work.
Sutro is a new AI-powered startup that allows users to build production-ready apps without coding experience in minutes. The platform automates aspects of app building, including AI expertise, product management, design, hosting, and scaling.
Founders can focus on their unique ideas while Sutro handles the technical aspects of app development, making the process as simple as creating a website.
The platform combines AI, including GPT-4, with rule-based compilers to generate web, iOS, and Android clients and set up the production back end. Users can make high-level changes and also enter their own custom code and integrations.
GitHub is launching an enterprise subscription tier for its Copilot pair-programmer that allows companies to customize it for their specific codebase.
The Copilot Chat feature, powered by OpenAI's GPT-4, will be available for general availability in December, and GitHub will roll out a new enterprise-grade Copilot subscription in February 2024.
GitHub is also introducing the Copilot Partner Program, allowing third-party developer tooling companies to build integrations for Copilot.
Mozilla has launched Fakespot Chat, an AI chatbot that helps online shoppers research products and answer questions about them. The chatbot uses AI and machine learning to analyze and sort product reviews to provide accurate information to users. The feature is available through the Fakespot Analyzer or as a browser extension for Amazon.com products.
Microsoft is partnering with venture capitalists to provide free access to its Azure cloud for startups to develop AI models.
The program, called Microsoft for Startups Founders Hub, will offer no-cost Azure AI infrastructure options for training and running generative models using GPU virtual machine clusters.
Y Combinator and its community of startup founders will be the first beneficiaries of this program, with plans to expand access to other venture funds and accelerators in the future.
Fabric is a new startup that offers an AI-powered service to organize documents and files, serving as a centralized workspace for information that can be queried using an AI assistant.
Users can create shared spaces within Fabric to collaborate on documents and chat, similar to Arc's shared folders and spaces.
Fabric uses AI technologies from OpenAI, Anthropic, and others to power its functionalities, including automatic speech recognition and file type detection.
OpenAI has introduced custom versions of ChatGPT called GPTs, which allow users to create tailored versions of the AI model for specific purposes.
GPTs can be used to provide assistance in various tasks such as learning board game rules, teaching math to children, or creating designs for stickers.
Initially, the ability to create GPTs is available for Plus and Enterprise users, with plans to offer this feature to more users in the future. Additionally, OpenAI will launch the GPT Store, allowing users to showcase and monetize their custom GPTs.
OpenAI has announced new tools that allow users to create custom chatbots and AI agents without coding skills. Users can specify what they would like the bot to do by chatting with OpenAI's ChatGPT, and the code needed to create the bot will be generated automatically. OpenAI will also launch an online chatbot store where users can find and share these custom bots.
The custom chatbots, called GPTs, can be used to help with specific tasks or interests, such as teaching math or designing stickers. Chatbot builders can monetize their creations by charging for access to their GPTs. Several companies, including Amgen, Bain, and Square, are already using GPTs internally.
OpenAI's new text-generation model, GPT-4 Turbo, is also introduced, which can process larger amounts of text and generate images and audio. The prices of all OpenAI's APIs have been reduced, making the cost of using the most advanced model more affordable.
LinkedIn is introducing AI tools powered by OpenAI to provide personalized digests and assist with content creation for its users.
The initial rollout of these tools will be for Premium users and will be available in three areas: feed customization, digesting linked articles, and improving the job-hunting experience.
LinkedIn is using OpenAI APIs, combining them with its proprietary data, to generate personalized AI outputs based on an individual's professional profile and activities on the site.
Google has launched Product Studio, a tool that uses generative AI to create product imagery for advertisers in the US. Merchants can input a prompt, such as a description or scene, and the AI model will generate a visual representation of the product based on that prompt. The tool can be used for simple tasks like changing the background color or for more complex requests like placing the product in a particular scene.
This AI-powered tool can improve low-quality images and remove distracting backgrounds, eliminating the need for reshoots.
Google has also added new features for merchants, including a "small business" attribute for highlighting businesses that have designated themselves as small businesses, and an enhanced knowledge panel that provides more information to shoppers when they search for a merchant's name.
Startup Freeplay emerges from stealth with $3.25 million in seed funding to help companies build and test apps powered by generative AI models, specifically text-generating models.
Freeplay aims to provide product development teams with tools to prototype and improve software features powered by large language models (LLMs), helping them adopt best practices and deliver better customer experiences.
The platform combines developer integrations with a web-based dashboard, offering observability, beginner-friendly features, and tools for custom evaluations of LLMs to optimize the customer experience and cut costs.
MassRobotics is launching a 13-week accelerator program for early-stage robotics startups, offering $100,000 in non-dilutive funding to accepted companies. The program will provide access to MassRobotics' facilities, including tools for hardware prototyping, and will feature human mentors for technical and business assistance. The application process is open until the end of November, with the program beginning in February 2020.
China's Vice Minister of Science and Technology, Wu Zhaohui, attended the AI safety summit in the UK and called for global cooperation and equal rights in accessing advanced AI.
Prominent Chinese computer scientist Andrew Yao joined Western academics to call for tighter controls on AI, warning of the potential existential risk it may pose to humanity.
China's participation in the event faced controversy, with former British Prime Minister Liz Truss warning of AI being used as a means of state control for Beijing. However, China has supported the UK's move and called for comprehensive discussions on global AI governance.
Yahoo spin-out Vespa has raised $31 million in funding from Blossom Capital to strengthen its engineering functions and deliver more features to its users. Vespa is an AI-powered big data serving engine that can handle large-scale datasets in real time, and is used by brands such as Spotify and OkCupid.
Vespa offers end-to-end services that allow clients to use a combination of text and structured data to provide relevant results at scale. It solves the problem of increasing customers and data for AI applications, while also allowing enterprise customers to leverage AI to streamline their operations.
Vespa, now an independent venture, has the capability to expand its cloud services and is encouraging its existing users to move to Vespa Cloud for managed services.
MIT researchers have developed a search engine called SecureLoop that can efficiently identify optimal designs for deep neural network accelerators, balancing data security and performance. The tool takes into account how encryption and authentication measures impact the energy usage and performance of the accelerator chip, enabling engineers to design secure and efficient accelerators tailored to their specific neural network and machine-learning tasks. When tested, SecureLoop identified schedules that were up to 33.2% faster and 50.2% more energy efficient than other methods that did not consider security.
Researchers from MIT and NVIDIA have developed two techniques that can accelerate the processing of sparse tensors, a type of data structure used for high-performance computing tasks. The first technique, called HighLight, efficiently finds and skips zero values in tensors, resulting in a six-fold improvement in energy efficiency. The second technique, called Tailors and Swiftiles, allows for overbooking the buffer occupancy, leading to more than double the speed and half the energy demands of existing hardware accelerators.
Kenyan B2B e-commerce company MarketForce is shutting down operations in Kenya, Nigeria, Rwanda, and Tanzania, and focusing its efforts on the Uganda market. The company is also launching a social commerce spinout called Chpter to help merchants utilize social media channels for increased sales.
MarketForce has experienced funding challenges and has shifted its focus towards profitability. The company raised $1 million through crowdfunding and is now shifting its resources to areas with high demand density. Uganda has been its best performing market due to exclusive distributor contracts and better margins.
MarketForce's RejaReja super-app, which allows informal retailers to order goods directly from manufacturers and access financing, will only be available in Uganda. The decision to exit the other markets was due to low margins, high competition, and expensive operations.
LAION, the nonprofit behind the Stable Diffusion text-to-image model, has launched the Open Empathic project, aimed at bringing emotion-detecting capabilities to open source AI systems. Volunteers can submit audio clips to a database to create AI models that understand human emotions, with the goal of making human-AI interactions more authentic and empathetic.
Emotion-detecting AI is being explored by various companies for purposes ranging from sales training to monitoring students' engagement in the classroom. However, accurately detecting emotions is challenging, as there are few universal markers and different cultures express emotions differently. The LAION team aims to address biases and work towards reliable emotion detection AI through community contributions.
While some advocates are calling for a ban on emotion recognition, LAION believes in the power of open development and transparency to ensure responsible use of AI technology. The organization welcomes suggestions and involvement from the community to make the Open Empathic project transparent and safe.
Outset is using GPT-4, OpenAI's text-generating AI model, to autonomously conduct and synthesize interviews with participants in research studies.
Outset allows researchers to create surveys and share the link with prospective survey takers, and GPT-4 follows up with respondents to gather deeper responses.
The company has already seen success with WeightWatchers, conducting and synthesizing over 100 interviews in 24 hours and using the results to propose a new framework for user segmentation.
Xpeng, the Chinese electric vehicle company, is removing high-definition mapping from its XNGP assisted driving feature, similar to Tesla's FSD.
While Tesla has eliminated HD maps and lidars, Xpeng still uses lidars but is now rolling out a map-free driving feature in 20 Chinese cities, with plans for 50 cities by the end of the year.
Other autonomous vehicle companies in China, like Deeproute, are also developing map-free autonomous driving solutions to reduce costs.
AI software developer Moreh has raised $22 million in a Series B funding round, with investors including AMD and Korean telco KT. Moreh's flagship software, MoAI, optimizes and creates AI models and is compatible with existing machine learning frameworks. The startup enables GPUs and other AI chips to operate AI models without code changes, and its performance has been shown to exceed that of Nvidia's DGX in terms of speed and GPU memory capacity.
KT has been working with Moreh since 2021 to develop a cost-effective, scalable AI infrastructure using AMD GPUs and MoAI software. Moreh aims to reach $30 million in revenue by the end of 2023 and plans to use the funding for research and development, product expansion, and hiring additional staff.
South Korean VC firms Smilegate Investment and Forest Partners, Moreh's existing investor, also participated in the Series B round.
Cisco has announced new AI tools for its Webex platform that aim to improve performance and automate meeting-related tasks.
The company has developed a real-time media model (RMM) that uses generative AI for audio and video to enhance the texture and context of meeting transcripts.
Cisco is also introducing an AI-powered audio codec that is up to 16 times more efficient in bandwidth usage and can recreate lost packets for crystal clear audio, even with significant packet loss.
The House Fund has closed its third tranche, Fund III, at $115 million, with the aim of investing in AI startups affiliated with UC Berkeley.
Ken Goldberg, a UC Berkeley professor, will join The House Fund as a part-time partner.
Fund III will primarily invest in pre-seed stage startups, but will also consider seed and Series A rounds. The House Fund has a focus on providing resources and support to entrepreneurs within the Berkeley community.
Direqt, a startup offering AI chatbot solutions for media companies, has raised $4.5 million in funding. The company provides customizable chatbot platforms for publishers like ESPN, GQ, Wired, Vogue, and others, allowing them to engage with their audience and monetize through ads. Direqt's platform supports various AI capabilities, including generative AI experiences, and can be integrated with messaging apps and social media platforms.
Publishers are increasingly interested in implementing generative AI experiences, with plans to do so in 2024, to improve engagement and traffic. Direqt's chatbot platform can scrape websites or leverage RSS feeds to ingest publishers' content, and the bots serve links to the content with a higher average clickthrough rate compared to emails. The company offers both a SaaS approach and a revenue-based model, allowing publishers to choose how they want to monetize their chatbots.
Google Play is implementing a new policy that requires developers to allow users to report offensive AI-generated content in Android apps.
The policy update is in response to issues with AI apps, such as apps that trick users into creating inappropriate imagery and those that enhance or alter images in potentially harmful ways.
The new policy will also review apps that request broad photo and video permissions and limit disruptive full-screen notifications to high-priority needs only.
Google has introduced a new feature called "About this image" to its image search results, which aims to provide users with more context and help them determine the reliability of an image. The feature will show when the image was first indexed by Google, its original source, where else it has appeared online, and whether it has been fact-checked. While it may not be a foolproof solution against misinformation, it is part of Google's ongoing efforts to combat the spread of fake or misleading media.
Amazon has introduced a new feature called "Explore with Alexa" that allows kids to have interactive conversations with a kid-friendly AI-powered Alexa. The content is generated offline, reviewed by a combination of AI and humans, and currently includes kid-friendly fun facts and trivia questions about animals. Amazon plans to expand the AI to cover other areas of interest to kids, such as space, music, video games, and sports.
Kids can access the AI-generated facts and trivia by speaking specific phrases to Alexa, or by engaging in organic conversations where the topic could come up. The AI experience is two-way, with Alexa also asking kids questions to engage them in the conversation.
Heidi Health, an AI platform, has raised $10 million in Series A funding led by Blackbird Ventures. The platform aims to alleviate the administrative burden on doctors by using AI to convert consultation transcripts into case histories and other documents. The platform also builds detailed clinical histories for providers and patients, improving the overall efficiency of healthcare visits.
Heidi Health's AI tools can prompt doctors to check for conditions and create clinical notes based on past visits. The platform also offers features such as My Additions, which allows clinicians to annotate transcripts during recordings, and a patient questionnaire to gather comprehensive health history.
YouTube Music is rolling out a new feature that allows users to create customized playlist art using generative AI, starting with English language users in the United States. Users can choose from a variety of themes and prompts, and the AI will generate a series of images to choose from. This feature aims to make it easier for users to express the uniqueness of their personal playlists.
YouTube Music is also planning to launch a new feature on the Home tab that will highlight users' recent favorites, making it easier to quickly jump back in and listen to songs and artists they love. This update is part of YouTube Music's effort to help users easily find and play their favorite music, similar to Spotify's functionality.
In addition to these new features, YouTube Music has recently introduced other functionalities such as a TikTok-style short-form video feed called "Samples" and timed lyrics. These updates aim to enhance the overall user experience and engagement on the platform.
D-ID has released a new mobile app that uses AI technology to turn still images into AI-generated videos. Users can upload an image and script, and the app will create a video with a digital representation of the person in the image. The app is aimed at AI enthusiasts and anyone who wants to create videos featuring digital people.
WorkMagic is an AI-powered platform that automates marketing tasks for small-scale Shopify sellers, saving them time and money.
The platform handles everything from photo and copy generation to campaign management and attribution analytics.
WorkMagic differentiates itself by offering automation of attribution, allowing users to measure the effectiveness of their campaigns and generate alternative content to improve results.
Amazon is set to begin testing Agility's bipedal robot, Digit, at its fulfillment centers, marking a potential milestone for the industry.
The move reflects Amazon's interest in exploring the potential of walking robots and how they can improve efficiency in warehouses and factories.
The outcome of the pilot test with Digit could shape the future of the humanoid robot industry, and its success could lead other companies to adopt similar robots.
Luzia, a Spain-based startup, has received $10 million in funding to expand its WhatsApp-based chatbot to the Spanish and Portuguese-speaking market.
The chatbot, called Luzia, has already attracted over 17 million users, with 8 million active on a monthly basis and receiving 13 million daily requests.
Luzia uses a combination of models, including GPT 3.5/4, Llama, and Kandinsky, and can generate text, transcribe voice notes, and even generate images based on prompts.
Amazon is testing Agility's bipedal robot, Digit, in its warehouses.
There is no guarantee that Amazon will deploy Digit to its warehouses, as testing is in the early stages.
Agility is one of several startups developing humanoid robots for warehouse work, and Amazon believes there is a big opportunity to scale Digit's capabilities to work collaboratively with employees.
Reality Defender, a startup developing tools to detect deepfakes and AI-generated content, has raised $15 million in a Series A funding round.
The funds will be used to expand the team and improve the AI content detection models.
Reality Defender offers an API and web app that analyze videos, audio, text, and images for signs of AI-driven modifications using proprietary models trained on in-house data sets.
Hook, a new app, aims to help users legally create remixes of their favorite songs for short video apps like TikTok.
Using AI technology, Hook allows users to pick short snippets of songs to remix and give their own spin, while compensating artists.
The app plans to launch in 2024 and has already raised $3 million in funding from investors.
Swiss startup Lakera has launched with the aim of protecting enterprises from security weaknesses in large language models (LLMs), such as prompt injections and data leakage. The company has developed a database of insights from various sources, including its own research and data collected from an interactive game called Gandalf. Lakera's flagship product, Lakera Guard, compares customer inputs against these insights to detect and prevent prompt injections and other types of cyberattacks.
Lakera is also focused on protecting companies from private or confidential data leaks, moderating content to ensure generative AI models don't serve up harmful or unsuitable content, and addressing inaccuracies or misinformation generated by LLMs. The company's launch comes at an opportune moment, as the EU AI Act is set to introduce regulations on LLM providers in the near future. Lakera's founders have served in advisory roles to the Act and aim to complement policy-making with developer-first perspectives.
Kobie Fuller, general partner at Upfront Ventures, is exploring the use of generative AI to reimagine the blog post and make it more interactive.
His idea involves turning a standard long-form blog post into various formats by creating a sophisticated AI conversation that can adapt to user queries.
Fuller has created a site called Kobie.ai to showcase the possibilities of this concept, using text-based conversational interfaces and AI simulated podcasts with AI interviewers.
Klarna has introduced a suite of new features, including an AI-powered image-search tool called Shopping lens, which allows users to take a picture of items and styles and find where to buy them. The company is also launching shoppable videos in Europe, in-store product scanning, a new cashback program, express refunds, and more. Klarna aims to become a place where consumers discover items and influencers promote products, expanding beyond a payments app.
AI chatbot startup, Character.AI, introduces a new group chat feature that allows users to chat with multiple AI characters at once.
Users can create group chats with only AI characters or a mix of humans and AI companions.
The feature is initially available to c.ai+ subscribers and will later be opened to the general public.
Google has unveiled its new AI-powered Assistant with Bard tool, and it looks like the Pixel 8 and Samsung Galaxy S24 phones will be the first to receive it.
Assistant with Bard will be available to "select testers" first, before rolling out to more users over the next few months.
After the Pixel 8 and Galaxy S24 owners have tested Assistant with Bard, the Pixel 6, Pixel 7, and Galaxy S23 handsets will likely receive the upgrade.
Box has announced a unique pricing plan for its AI functionality, which uses a credit system. Each user gets 20 credits per month, and if they go over that, they can dip into a shared pool of 2000 additional credits belonging to the entire company. Customers can also buy additional blocks of credit for extended usage.
The pricing plan was designed to balance the cost of running the generative AI model with the need for fair pricing for customers and recognizing higher volume usage by power users.
The first two Box AI features, creating content in Box Notes and asking questions about specific documents, will be available in beta for Enterprise Plus subscribers next month, operating under the new pricing model.
AMD has acquired Nod.ai, an open-source AI software provider, to enhance its ability to provide customers with AI models tailored for AMD hardware.
The acquisition is expected to be completed in this quarter, but no financial details have been disclosed.
AMD aims to advance open-source compiler technology and deliver high-performance AI solutions across its product portfolio with the addition of Nod.ai's talented team.
The UK’s data protection watchdog, the ICO, has raised concerns about Snap’s AI chatbot ‘My AI,’ specifically regarding children’s privacy risks. The regulator stated that Snap’s risk assessment before launching the chatbot did not adequately assess the data protection risks posed by the generative AI technology, particularly to children. Snap will have the chance to respond before the ICO makes a final decision on whether the company has breached data protection rules.
Snap launched the generative AI chatbot, powered by OpenAI’s ChatGPT, in February. The chatbot was initially available only to subscribers but later opened to free users. Despite moderation and safeguarding features in place, there have been reports of the chatbot providing inappropriate advice. Some users have also bullied the AI.
European privacy regulators have previously scrutinized AI chatbots, including Italy’s Garante and Poland’s data protection authority investigating similar concerns.
Venture firm Section 32 has raised $525 million for its fifth fund, bringing its total assets under management to $2.3 billion. The firm invests in software-driven businesses in tech and healthcare, with a focus on areas such as infrastructure, cybersecurity, gaming, and precision medicine. Section 32 is cautious about investing in AI, as it believes that certain capabilities may become commoditized and offered by big companies for free or through existing software subscriptions.
Google Photos' Magic Editor feature will be available on the Pixel 8 and 8 Pro smartphones, using generative AI to perform complex photo edits like removing objects or repositioning subjects.
The Magic Editor feature combines various editing tasks, allowing users to easily edit photos without the need for other tools or manual edits.
Google Photos is leveraging generative AI to make these edits, offering multiple output options for users to choose from.
Meta has introduced new generative AI features for advertisers, including the ability to create custom backgrounds, adjust images to fit different aspect ratios, and generate multiple variations of ad text based on the original copy. These tools aim to assist brands and businesses in delivering effective advertising campaigns and have shown promising early results in saving advertisers time. Meta also plans to develop more AI features, such as generating ad copy and backgrounds with tailored themes.
LinkedIn is launching new AI features across its job hunting, marketing, and sales products, including an AI-powered LinkedIn Learning coach and an AI-powered tool for marketing campaigns.
The platform is leveraging technology from OpenAI and Microsoft to power these new features.
The AI-assisted recruiting experience, called Recruiter 2024, will use generative AI to help recruiters come up with better search strings and provide search suggestions.
Visa plans to invest $100 million in generative AI companies that are developing technologies and applications for the future of commerce and payments.
Visa Ventures will make the investments, looking for companies using generative AI to solve problems in commerce, payments, and fintech.
Visa is interested in companies at various levels of the stack, from data organization for generative AI to user experiences, and is particularly interested in responsible use of AI.
OpenAI has announced the release of GPT-4V, which incorporates image analysis capabilities into the GPT-4 language model.
The integration of image inputs into large language models is seen as a significant development in AI research.
OpenAI has conducted evaluations and implemented safety measures specifically for GPT-4V's image analysis capabilities.
New voice capabilities are being added to ChatGPT, allowing users to engage in voice conversations with their assistant, request stories, and settle debates.
Users can now use image input with ChatGPT to troubleshoot issues, plan meals, or analyze complex data. They can also use the drawing tool to focus on specific parts of the image.
Voice (Beta) is currently available for Plus users on iOS and Android, while image input is generally available for Plus users on all platforms.
The MIT and Accenture Convergence Initiative for Industry and Technology has selected five new research fellows for 2023-24. The fellows will conduct research in areas including artificial intelligence, sustainability, and robotics.
Yiyue Luo, a PhD candidate, will research and develop novel sensing and actuation devices using digital manufacturing and AI, with the goal of revolutionizing interactions between people and their environments.
Zanele Munyikwa, a PhD candidate, will focus on the impact of foundation models on work and tasks in industries such as marketing, legal services, and medicine, as well as explore the convergence of creative and technological industries enabled by foundation models.
OpenAI is opening a new office in Dublin, Ireland to support the growth of artificial intelligence in Europe.
The company plans to collaborate with the Irish government and industry to advance AI development and deployment.
OpenAI's investment in Ireland is seen as an endorsement of the country's focus on building a flourishing AI ecosystem.
1. The article discusses the use of AI in various industries and how it has significantly impacted them.
2. It highlights the potential of AI to automate tasks, improve efficiency, and enhance decision-making processes.
3. The article also mentions the ethical concerns and challenges associated with AI, including privacy and job displacement.
OpenAI is launching ChatGPT Enterprise, which offers advanced security and privacy features, unlimited access to GPT-4, longer context windows, customization options, and advanced data analysis capabilities.
ChatGPT Enterprise has seen widespread adoption in organizations, with over 80% of Fortune 500 companies using it. Early users have reported using it for tasks such as clear communication, coding tasks, exploring complex business questions, and assisting with creative work.
ChatGPT Enterprise provides enterprise-grade security and privacy, with data owned and controlled by the customer. It also offers unlimited access to GPT-4, faster performance, advanced data analysis, and collaboration features with shared chat templates.
The MIT and Accenture Convergence Initiative for Industry and Technology has selected three new research projects that focus on sustainability, digital health, and the future of work. The projects aim to accelerate progress in meeting complex societal needs through new business convergence insights in technology and innovation.
The first project, led by Jessika Trancik, aims to identify how industrial clusters can enable companies to derive greater value from decarbonization to address climate change. The second project, led by Anette Hosoi, will develop a return-on-investment (ROI) calculator for childhood obesity interventions to encourage companies to invest in reducing childhood obesity. The third project, led by Thomas Malone, aims to use natural language processing algorithms to reshape the future of work by better matching applicants to jobs and identifying skill training needs.
The selected research projects have the potential to have a tremendous impact and to guide and shape future innovations in sustainability, digital health, and the future of work.
OpenAI and Scale are partnering to offer fine-tuning capabilities to companies using their advanced AI models, starting with GPT-3.5 Turbo and soon expanding to GPT-4.
Fine-tuning allows companies to customize OpenAI's models with their proprietary data, making them more powerful and useful.
Scale, as a preferred partner, will provide enterprise AI expertise and data enrichment services to help customers effectively leverage the fine-tuning capability.
The MIT Schwarzman College of Computing has awarded seed grants to seven interdisciplinary projects exploring how artificial intelligence and human-computer interaction can be used to enhance management and productivity in modern workspaces.
The projects bring together researchers from computing, social sciences, and management to conduct research in this rapidly evolving area.
The selected projects include designing memory prosthetics using large language models, simulating social scenarios with AI agents, exploring the impact of AI on human decision-making, studying how generative AI can improve job quality in healthcare settings, democratizing programming with generative AI tools, understanding the impact of AI on skill acquisition and productivity, and developing AI-powered onboarding and support systems.
ChatGPT users on the free plan can now access custom instructions, with availability in the EU & UK coming soon.
Users can personalize their interactions with ChatGPT by providing specific details and guidelines for their chats.
To add custom instructions, users can click on their name and select 'Custom instructions'.
Researchers at MIT's Computer Science and Artificial Intelligence Laboratory (CSAIL) have developed a technique called "PhotoGuard" that uses invisible perturbations to disrupt an AI model's ability to manipulate images. The technique involves two attack methods: an "encoder" attack that adjusts the image's latent representation to make it appear random, and a "diffusion" attack that optimizes perturbations to resemble a target image. Although PhotoGuard is not foolproof, it offers a preemptive measure to protect images from unauthorized manipulation by AI models.
Researchers from MIT and Stanford University have developed a machine-learning technique that can efficiently learn to control a robot in dynamic environments with rapidly changing conditions. This technique uses structure from control theory to learn a model that can extract an effective controller directly from the model, reducing the need for separate controller learning. The researchers’ approach is able to learn an effective controller using fewer data and achieves better performance compared to other methods.
The technique can be applied to various types of robots, such as drones and autonomous vehicles, and enables them to navigate challenging conditions, such as slippery roads or strong winds. The researchers’ approach incorporates structure in the dynamics of the system, which helps guide the control logic and leads to more effective, stabilizing controllers.
The new machine-learning method is data-efficient, achieving high performance with fewer data points. It outperforms other approaches that require separate learning of dynamics and controller. The technique is inspired by how roboticists use physics to derive simpler models for robots and combines learning dynamics with control-oriented structure from data to create more efficient controllers.
Anthropic, Google, Microsoft, and OpenAI have launched the Frontier Model Forum, an industry body focused on ensuring the safe and responsible development of frontier AI models. The Forum aims to advance AI safety research, identify best practices for frontier models, share knowledge with policymakers and the community, and support efforts to leverage AI for societal challenges. The Advisory Board will guide the Forum's strategy and priorities.
The Frontier Model Forum will focus on identifying best practices, advancing AI safety research, and facilitating information sharing among companies and governments. They will promote knowledge sharing, establish research priorities, and develop secure mechanisms for sharing information on AI safety and risks.
The Forum plans to collaborate with civil society, governments, and existing initiatives such as the G7 Hiroshima process, the OECD's work on AI risks, and MLCommons. They will establish key institutional arrangements, consult with stakeholders, and support ongoing efforts in the AI community.
ChatGPT for Android is now accessible in the United States, India, Bangladesh, and Brazil via the Google Play Store.
The availability will soon be extended to more countries within the next week.
Users can stay updated on the Android rollout through a tracking mechanism provided.
The American Journalism Project (AJP) has announced a partnership with OpenAI to explore ways in which artificial intelligence (AI) can support local news. OpenAI is committing $5 million to AJP to support its work and providing up to $5 million in API credits to help organizations deploy AI technologies. The collaboration aims to develop tools that can assist local news organizations and address challenges such as misinformation and bias.
The partnership will include the creation of a technology and AI studio, which will assess the applications of AI within the local news sector. AJP will also distribute grants to approximately ten of its portfolio organizations to pilot and experiment with various AI applications. Additionally, OpenAI will provide API credits to AJP and its portfolio organizations to build and utilize AI-powered tools.
The American Journalism Project is dedicated to addressing the market failure in local news and has raised $139 million to support nonprofit local news organizations. OpenAI, founded in 2015, is focused on ensuring that AI benefits all of humanity.
The browsing beta feature can sometimes display content in unintended ways, such as fulfilling requests for a URL's full text when not intended.
To address this issue, the browsing beta feature will be temporarily disabled.
The browsing beta feature will be fixed to ensure that it displays content correctly and fulfills user requests as intended.
OpenAI has opened its first international office in London to expand its operations and accelerate its mission of ensuring that artificial general intelligence (AGI) benefits humanity.
The London office will focus on advancing OpenAI's research and engineering capabilities in AGI development and policy.
This expansion provides an opportunity for OpenAI to attract world-class talent and drive innovation while collaborating with local communities and policy makers.
The National AI Advisory Committee will establish a subcommittee focused on AI in law enforcement, providing advice to the President on topics including bias, data security, adoptability of AI, and legal standards. The subcommittee includes experts such as Assistant Chief Armando Aguilar, who has implemented offender-focused strategies and developed facial recognition technology policy for the Miami Police Department. Other members include Anthony Bak, Head of AI for Palantir, and Amanda Ballantyne, Director of the AFL-CIO Technology Institute.
The NAIAC consists of leaders from across academia, non-profits, civil society, and the private sector with expertise in AI. They provide advice on topics like research, ethics, governance, and technology transfer. Chairperson Miriam Vogel is the CEO of EqualAI, focusing on reducing bias in AI and promoting responsible governance.
James Manyika, the Vice Chair, is Senior Vice President for Technology & Society at Google and leads Google Research. He has extensive experience in technology and the economy, having served in various leadership roles in government and academia.
Other members include Yll Bajraktari, CEO of the Special Competitive Studies Project, Amanda Ballantyne, Director of the AFL-CIO Technology Institute, and Sayan Chakraborty, co-president of Workday's product and technology organization.
Note: Additionally, users now have the option to continue generating messages beyond the maximum token limit, with each continuation counting towards the message allowance.
Experimental support for AI plugins in ChatGPT is being introduced, allowing the use of tools designed for language models.
Plugins can be used to access current information, perform computations, or utilize third-party services.
The initial set of plugins being rolled out includes Browsing, Code Interpreter, and Third-party plugins.
1. Researchers have trained a neural network to play Minecraft using a method called Video PreTraining (VPT) that utilizes unlabeled video data. The model can perform tasks such as crafting diamond tools, which typically takes humans over 20 minutes to complete.
2. The VPT method involves training an inverse dynamics model (IDM) using a small labeled dataset and then using the IDM to label a larger dataset of online videos. This allows the model to learn to act via behavioral cloning.
3. The researchers fine-tuned the VPT model using a specific dataset to improve its performance in building houses and performing early game skills in Minecraft, such as crafting tools and constructing shelters.