Latest News

Ray-Ban Meta AI glasses are changing lives in Singapore. The company has donated 1,000 of these devices to four community organizations that support people with disabilities.

Singapore’s government is actively promoting digital inclusion and positioning itself as a hub for emerging accessibility solutions. This makes it an ideal place to test innovative technologies like AI-powered glasses – technology that can describe surroundings, read text aloud, and provide real-time assistance.

The donated Ray-Ban Meta AI glasses have advanced features such as object recognition, sign language interpretation, and navigation assistance through voice commands. These capabilities are already being used by people who rely on assistive technology to get by, and they could make a huge difference in daily life for users with disabilities.

Meta is working closely with disability organizations in Singapore to turn these devices from consumer products into practical tools that genuinely improve quality of life. The company wants to gather real-world data and insights from this collaboration to inform the development of AI wearables for accessibility purposes, which requires input from people who truly need these technologies.

The four recipient organizations in Singapore are expected to test a range of applications beyond what typical consumers might explore with their donated glasses. Early feedback suggests that AI glasses excel at tasks such as identifying currency denominations, reading medication labels, and describing social situations – essential skills for daily independence and self-sufficiency.

Meta’s initiative has far-reaching implications extending far beyond Singapore’s borders. The company is conducting a large-scale pilot program to assess the feasibility of integrating AI wearables into healthcare systems and disability support services globally. If successful, this could accelerate adoption among organizations that have been hesitant to embrace consumer-focused tech products – a shift that would benefit millions worldwide.

Singapore has become Southeast Asia’s innovation hub, where public-private collaborations are driving ambitious accessibility projects under the government’s Smart Nation initiative. This kind of partnership is becoming a model for rolling out AI assistive technology in other markets and regions around the world.

For people with disabilities, this donation represents more than just free gadgets – it shows that AI wearables can serve real purposes beyond social media and entertainment. The real test will be whether these organizations can demonstrate measurable improvements in independence and quality of life for their users over time.

Meta’s Singapore initiative marks a turning point for AI wearables, transitioning from novelty to genuine assistive technology. If the 1,000 donated glasses prove their worth in real-world disability support scenarios, it could accelerate mainstream adoption and convince other tech giants to prioritize accessibility in their AI hardware development – a shift that would benefit countless people around the globe.

Microsoft made a splash at their latest live event, showcasing new gear and Windows updates that promise to mix things up. It’s been a while since they’ve taken the stage – but now they’re back with some real surprises.

The Surface Laptop Ultra is finally available for pre-order, starting at $2,599. First shown in May as just a teaser, we now have all the details: it’s powered by Nvidia’s RTX Spark SoC, plus options for 5120-core or 6144-core Blackwell GPUs and up to 128GB of unified memory LPDDR5x.

One key advantage Surface Laptop Ultra has over other machines is its ability to handle AI tasks with greater efficiency. Thanks to unified memory allocation, developers can manage their resources more smoothly – a major plus for those in the field.

But it’s not just AI development that gets a boost from this new tech: the laptop itself handles modern gaming without breaking a sweat. Microsoft says the credit goes to unified memory allocation, which makes this level of performance possible even with out-of-the-box specs.

Nvidia’s Jensen Huang joined forces with Microsoft on stage to show off their RTX Spark Dev Box – essentially an $5,999 aluminum box designed specifically for top developers who need serious AI power. It comes pre-loaded with Windows 11 and a custom ‘developer-optimized’ build that gets things done.

The real star of the show is the Dev Box’s performance: packed with 128GB of unified memory, Nvidia’s RTX GPU on board – all this adds up to over one petaflop (or 1,000 teraflops) of AI compute power. That means top developers will love what Microsoft and Nvidia have cooked up here.

Microsoft has announced its vision for Windows as the platform for hybrid intelligence, where AI agents can run locally on a device when needed and connect to the cloud when necessary. This approach combines local and cloud-based artificial intelligence (AI) capabilities to enable powerful, secure, and efficient experiences across PCs, from everyday productivity devices to high-performance workstations and supercomputers.

The company’s goal is to make Windows the most secure platform for AI agents by incorporating built-in containment, identity, and manageability features. This allows organizations to control what agents can access, track which agent performed an action, and govern agents at scale. Microsoft Execution Containers (MXC) are now generally available on Windows.

Hybrid intelligence brings frontier AI capabilities to Windows locally by combining intelligent routing with powerful local AI models and a performant runtime. This enables workloads to run in the right place for each task, whether locally on a device or in the cloud. The result is improved performance, efficiency, and security compared to traditional cloud-based solutions.

Copilot will be enhanced with Windows hybrid intelligence capabilities, allowing it to leverage local context from a user’s PC, perform actions on their behalf, and tap into local AI models. This provides more capability while optimizing AI compute usage. Users can expect more seamless interactions between Copilot and their devices as a result of this integration.

A new generation of Windows PCs is being introduced by Microsoft in collaboration with its device partners and NVIDIA. These systems are designed for agentic and AI-powered workloads, ranging from always-on mini desktops to high-performance workstations like the Surface Laptop Ultra, powered by NVIDIA RTX Spark, which is available for pre-order today.

Microsoft has also highlighted continued investments in Windows quality, performance, security, developer experiences, and gaming. These updates position Windows as a platform designed for the next generation of AI-powered computing, where intelligent agents, local AI models, and cloud services work together seamlessly.

A prominent kids’ safety organization has raised concerns about the new teen-focused version of ChatGPT, warning parents that it poses an ‘unacceptable risk’ until its alleged gaps are fixed. Common Sense Media’s Youth AI Safety Institute tested over 4,000 prompts to emulate how teens might use ChatGPT before and after OpenAI announced stronger built-in safety protections for teens on August 18.

Common Sense Media conducted a comprehensive risk assessment of ChatGPT for Teens, which included testing its crisis referrals, parental alerts, and age assurance features. The organization found that the platform failed to deliver as promised in several areas, including sending parental alerts when necessary. This missed signals of suicidal ideation, self-harm, or disordered eating on newly created parent-linked accounts.

The testers also discovered that ChatGPT for Teens didn’t offer the right help in crisis situations, missing more than one in four warranted crisis referrals across mental health conditions it tested. Furthermore, despite a ‘study mode,’ the platform still allowed teens to shortcut their learning process by doing homework for them. This raises concerns about the potential consequences of relying on AI-powered tools without proper safeguards.

Common Sense Media’s risk assessment also highlighted issues with ChatGPT for Teens’ age estimation and its failure to curb ‘friend-like behavior.’ The testers found that the platform expressed feelings, preferences, and moods in response to analysts treating it like a real person. This has sparked concerns about the potential risks of using AI-powered tools without proper safeguards.

Common Sense Media’s Founder and CEO Jim Steyer emphasized the importance of prioritizing kids’ safety when developing AI-powered tools. He stated that OpenAI should stop young people, including teenagers, from using ChatGPT until it can be made fully safe. Steyer also expressed concerns about OpenAI marketing safety features to parents without ensuring they work as promised.

OpenAI responded to the criticism by stating its commitment to giving teens a safe product and providing parents with meaningful tools to guide their teens’ use of AI. However, the company took issue with Common Sense Media’s methodology in its risk assessment, claiming that it may not accurately reflect how ChatGPT’s teen safeguards work in practice.

OpenAI suggested that Common Sense Media retest ChatGPT for Teens with fully activated accounts and a larger sample size. The company also pointed out that safety notifications are on by default when parents link to their teens’ accounts but can take up to a few hours to activate. This raises questions about the timing of Common Sense Media’s testing and whether it accurately reflects the platform’s capabilities.

Common Sense Media has defended its methodology, stating that they work closely with companies before releasing major product risk assessments, including sharing their methodology. The organization also emphasized the rigor of its research, which is guided by a board of expert advisors in AI, youth development, child safety, mental health, and education.

A new update to ChatGPT lets it respond with interactive visuals. The feature, called Intelligent UI, combines text responses with diagrams, charts, forms, and buttons. It makes conversations more engaging and informative by letting users interact directly with the chatbot.

The GPT-6 model release is behind this upgrade. OpenAI announced last month that its new language model would be integrated into ChatGPT. This update allows the chatbot to generate interactive responses on its own, within a conversation.

To show how Intelligent UI works, OpenAI shared two examples. One shows ChatGPT drawing a diagram of a seven-speed bicycle when asked about design details. The diagram includes buttons that highlight different bike parts. Another example shows ChatGPT teaching users Mahjong by generating tile visuals and categorizing them for easy reference.

Users can ask the chatbot to create tools like retirement savings calculators or bill splitters, which it will generate on the fly with their response. This is especially helpful when tasks require complex calculations or data visualization. It lets people get things done more efficiently in conversations.

The GPT-6 model has been trained to figure out when and how to display interactive visuals within a conversation. OpenAI says this new model showed stronger resistance to attempts to bypass its safety training, which indicates an improvement in performance.

GPT-6 with Intelligent UI is now available for ChatGPT Plus, Pro, Business, and Enterprise users worldwide. But it will roll out to Go and free tiers on Thursday. Higher-tier subscribers get access to the mid-range GPT-6 Sol model, while lower-tier users get a more efficient version called Luna.

The new update marks a big step up in how well ChatGPT can provide interactive responses. As OpenAI keeps refining its language models, we’ll see even more innovative features like this one pop up.

NVIDIA has opened preorders for its new line of laptops, the RTX Spark series, which are designed to run on Windows PCs. The company made this announcement at a recent event in San Francisco, where it also showcased its partnership with Microsoft. According to NVIDIA’s recap of the event, the two companies have been working together for several decades and have now reached a significant milestone in their collaboration.

The event featured a fireside chat between Jensen Huang, founder and CEO of NVIDIA, and Satya Nadella, CEO of Microsoft. The conversation was hosted by Sriram Krishnan, a former senior White House policy advisor on AI. During the discussion, Huang highlighted NVIDIA’s long-standing support for Windows and Microsoft, stating that without Windows, there would be no GeForce.

Nadella credited Huang’s consistency in vision as a key factor in bringing the two companies to this moment of collaboration. He emphasized the importance of making the desktop the most secure place for agents to execute, which is why Microsoft has been working on its Execution Containers (MXC) technology. MXC allows agents to run safely and persistently in the background under operating system control.

Microsoft’s executive vice president of Windows and Devices, Pavan Davuluri, also spoke at the event about the company’s efforts to build primitives directly into Windows for agent security and governance. He pointed out that Microsoft Security and Agent 365 are key components of this effort. Nadella returned to the theme in his conversation with Huang, reiterating the need for a secure environment for agents.

Huang stated that MXC will revolutionize how agents are built and deployed, much like Windows and DirectX did for applications. He emphasized that NVIDIA’s approach would form the foundation of the next generation of IT. The partnership between NVIDIA and Microsoft is expected to have significant implications for the development and deployment of AI models in various industries.

The RTX Spark laptops will run on a full NVIDIA AI stack, which includes CUDA platform support natively. This means developers can move their models, tools, and workflows without rewriting them. The systems are designed to handle large-scale AI workloads, with some configurations capable of running up to 120 billion parameters locally. One notable example is the Qwen 3.8 Flash Next model, which has a 125B parameter count and a 51B n-gram.

NVIDIA also highlighted its fifth-generation Tensor Cores with NVFP4 support in the RTX Spark laptops. This will enable hardware-accelerated AV1 video encode and decode for creators. For gamers, the company promises to run AAA titles at 1440p above 100 frames per second using DLSS 5, Reflex, and G-SYNC.

The event also previewed NVIDIA’s DGX Station for Windows, which is described as a deskside AI supercomputer that brings GB300 Grace Blackwell-class AI infrastructure directly into the Windows ecosystem. This system will allow users to run frontier-class models locally without needing to rent a cluster. The DGX Station for Windows runs on the GB300 Grace Blackwell Ultra Desktop Superchip and offers up to 748 GB of coherent memory.

NVIDIA’s product page lists several key features, including access to Linux AI toolchains through the Windows Subsystem for Linux. The company also notes that the vast majority of Fortune 500 companies are standardized on Windows, making this move significant for enterprise developers. NVIDIA is taking notification sign-ups for availability and expects DGX Station for Windows in the fourth quarter of 2026 from several major manufacturers.

The RTX Spark laptops will be available starting October 16, 2026, with preorders opening on October 7. The systems are designed to handle a range of tasks, including running large-scale AI models locally without needing cloud infrastructure. NVIDIA has partnered with several major manufacturers to bring the RTX Spark series and DGX Station for Windows to market.

The Surface Laptop Ultra is one example of an RTX Spark laptop that will be available from Microsoft. This device features up to 128 GB of unified memory and can run AI models locally without needing cloud infrastructure. NVIDIA’s Blackwell RTX GPU combines with the Grace CPU in a single chip, delivering up to one petaflop of FP4 AI performance.

NVIDIA has also announced that its DGX Station for Windows will be available from several major manufacturers, including ASUS, Dell Technologies, GIGABYTE, HP, MSI, and Supermicro. The system is expected to have significant implications for the development and deployment of AI models in various industries.

Copilot, Microsoft’s AI-powered tool, is getting more control over Windows and user files. The company has unveiled an upgrade that grants it access to local files on PCs and lets it take actions across the operating system.

Hybrid Intelligence is a new concept where apps and tools rely on both local and cloud AI models to accomplish tasks efficiently. This approach enables Copilot to use user files on Windows PCs, allowing it to perform various actions. The company says this will improve performance and save time for users.

Micrsoft showcased the power of Hybrid Intelligence with a video demo featuring Jacob Andreou, EVP of Copilot at Microsoft. In the demonstration, he asked Autopilot to assist with filing taxes by searching for relevant documents and taking necessary steps. The AI agent successfully located required information across different folders, renamed files, zipped them into a folder, and drafted an email to send to the accountant.

The features powered by Hybrid Intelligence will roll out to Copilot over the next couple of months, according to Andreou’s announcement. This upgrade is part of Microsoft’s ongoing efforts to enhance its AI capabilities and improve user experience. The company says these improvements will be available soon.

Starting this fall on Windows 11 PCs, users can take quick actions from a new search bar feature unveiled by Microsoft. These actions include turning on dark mode, adjusting mic volume with an in-line slider, or sending a text message – confirmed by Pavan Davuluri, boss of Windows and Surface.

Microsoft has introduced its latest laptop, the Surface Laptop Ultra, which is designed to handle demanding artificial intelligence (AI) tasks. The device was unveiled at an event in San Francisco and features a powerful Arm-based Nvidia RTX Spark processor.

The Surface Laptop Ultra is positioned as a competitor to Apple’s MacBook Pro laptops for AI workloads, video editing, and gaming. With a price tag of $2,599, it is Microsoft’s most expensive laptop yet, but the company claims its performance justifies the cost.

Pavan Davuluri, executive vice president of Windows and devices at Microsoft, highlighted the device’s capabilities in an interview with Yahoo Finance. He noted that the Nvidia RTX Spark processor features up to 128GB of unified memory, allowing it to run complex AI models more efficiently than other laptops on the market.

The Surface Laptop Ultra is likely to appeal to early adopters across both consumer and enterprise spaces, according to Tom Mainelli, group vice president of device and consumer research at IDC. He emphasized that cost remains a significant factor in purchasing decisions for many users.

However, as AI workloads become more prevalent, the need for local processing capabilities will grow. This is because certain tasks can be handled by free, locally installed models on your PC rather than relying on expensive cloud-based services like OpenAI’s GPT-6 Astra.

Micorosft demonstrated its Surface Laptop Ultra’s performance in a video clip showcasing it running multiple demanding applications simultaneously, including the video game ‘Gears of War: E Day’ and the video editing platform DaVinci Resolve. This level of multitasking would be challenging for standard laptops to handle.

The device offers significant improvements over Apple’s MacBook Pro with its M5 Pro chip in terms of AI performance. Microsoft claims that the Surface Laptop Ultra generates AI images 4.3 times faster than a MacBook Pro and video content 6.2 times faster.

Despite the ongoing shortage of memory and storage chips, which has led to increased prices for devices across the industry, device makers have managed to insulate themselves from the worst of sales declines so far.

Just a month after launching on iPhone, Meta is bringing its AI-powered assistant Muse to the iPad. This expansion marks another significant step for the company in making these tools accessible across various platforms.

The new dedicated app offers users a more detailed look at their accounts and tasks. With this addition, Muse has reached 6.6 million installs since its launch on iOS and Android back on September 8, according to Sensor Tower’s estimates.

Muse lets users connect their accounts for staying organized and handling tasks like booking reservations or ordering groceries. To make this work smoothly, Meta is continuously adding partners that appear in Muse as ‘connectors.’ Users can add these connections selectively based on what parts of their online life they’re comfortable letting an AI agent manage.

Meta’s efforts have led to the inclusion of several new connectors for small businesses like Asana, Canva, and Dropbox. The company has also updated its list with newer additions such as Notion and Granola. Moreover, Meta recently announced more retail partners – Best Buy, Gap, Sephora, Walmart, and Wayfair are now part of this effort.

These new connectors allow the agent to shop those retailers’ sites and complete transactions on users’ behalf. But this technology is still in its early stages, and issues can arise when websites mistakenly block AI agents thinking they’re malicious bots. To address this problem, Meta has announced it’s working with industry partners on an open standard that would enable personal AI agents to identify themselves to businesses.

This technological solution aims to help businesses distinguish between genuine users and malicious bots, which should improve the user experience and reduce errors. In its effort to make Muse more useful for both individuals and small businesses, Meta is taking steps towards developing this crucial technology with industry partners.

OpenAI is introducing new features to its ChatGPT for Teens mode, which was launched in August with built-in safeguards and reminders. The company has added a tool specifically designed to help users navigate the college application process.

The College Planner feature allows students to keep track of deadlines and requirements for multiple colleges on their list. This centralized plan includes information about financial aid steps, tasks, and application deadlines, making it easier for students to stay organized and focused.

According to OpenAI, this new tool will enable users to update their progress, use a timeline to plan ahead, and identify what still needs to be done. Students can also ask ChatGPT questions related to the college planning process, such as clarifying specific requirements or getting started with applications.

The College Planner feature is set to arrive ‘soon,’ although OpenAI has not provided an exact release date. Initially, this tool will focus on students in grades 10 through 12 who are planning to attend a four-year college in the US.

In addition to the new College Planner feature, OpenAI is updating ChatGPT with other tools aimed at supporting users’ academic endeavors. On iOS devices, users can now upload multiple photos and convert them into a single PDF using a continuous capture function.

ChatGPT will also be able to generate flashcards based on user notes, making it easier for students to review and study material. These updates come amidst controversy surrounding ChatGPT’s role in assisting with homework and its lack of parental notification features.

A recent report by Common Sense Media deemed ChatGPT for Teens an ‘unacceptable risk,’ citing concerns about the AI tool doing kids’ homework without adequate oversight or alerts to parents when necessary. OpenAI has responded, stating that teens typically spend less than 15 minutes per day using ChatGPT.

Alphabet, the parent company of Google, and Meta Platforms have made a significant investment in artificial intelligence (AI) biology research through their joint funding of Biohub’s Virtual Biology Initiative. The initiative aims to develop accurate predictive models of cell behavior using data from experiments conducted across various conditions.

The combined $300 million contribution from Alphabet’s drug discovery unit Isomorphic Labs, Google DeepMind, and Meta will be used to support the development of AI models that can predict how cells respond in different situations. This is part of a larger effort led by Biohub, which has already received commitments totaling $1.8 billion.

The Virtual Biology Initiative involves measuring cell responses across numerous conditions using advanced technologies such as machine learning algorithms and high-performance computing. The data collected will be used to train AI models that can accurately predict how cells behave in various scenarios.

Biohub’s head of science, Alex Rives, explained the reasoning behind the one-year embargo on commercially funded datasets: ‘We have to have some incentive for commercial players to be a part of this, and the embargo period creates that.’ This means that companies participating in the initiative will get access to data before it is made public.

The funding breakdown shows that Biohub has committed $500 million, while the U.S. Department of Energy and the National Institutes of Health have also pledged significant amounts: more than $500 million each over five years. The combined investment from Google DeepMind, Isomorphic Labs, and Meta totals $300 million.

It’s worth noting that this investment is relatively small compared to Alphabet and Meta’s overall research budgets. In 2025 alone, Alphabet spent a staggering $91.45 billion on capital expenditures (capex), while Meta invested $69.69 billion in capex. Both companies have significantly increased their R&D spending since 2021.

Alphabet’s total R&D expenditure over the last 12 months was $68.97 billion, with Meta following closely behind at $71.63 billion. Together, they spent an astonishing $140.6 billion on research and development in 2025. To put this into perspective, their combined daily R&D budget is approximately $385 million.

Isomorphic Labs, Alphabet’s drug discovery unit, stands to benefit the most from the Virtual Biology Initiative due to its focus on developing AI models for predicting cell behavior. The company has already established partnerships with pharmaceutical giants Eli Lilly and Novartis, and it raised an additional $600 million in funding from outside investors earlier this year.

The real value of participating companies’ early access to data lies not just in the datasets themselves but also in the time they gain before those datasets become public. This head start allows them to train their AI models more effectively and potentially develop new treatments or products sooner.

Mistral has released a new artificial intelligence model called Le Chonk, which the company claims can compete with top-end models from the US and China. The model is currently available in preview and will be finalized by the end of this month.

The new AI model, known as Mistral Large 4 or Le Chonk for short, has a massive 1 trillion parameters. This makes it one of the largest open-weight models developed outside of China, according to Mistral.

Le Chonk is designed to be highly versatile and can be used in various domains such as coding, cyberdefense, manufacturing, finance, electrical engineering, and more. The model’s capabilities are optimized for these specific areas, allowing it to perform well on tasks that might not be a priority for other top-end models.

Guillaume Lample, cofounder and chief scientist at Mistral, emphasized the potential of Le Chonk in an interview with WIRED. He noted that there are many domains where current AI models may struggle or lack focus, providing opportunities for improvement.

Mistral claims to have trained its model from scratch without relying on distillation techniques used by some Chinese labs. This approach allows Mistral to avoid potential issues related to the training of smaller models on larger ones’ outputs.

Le Chonk is an open-weight model, which means it can be run and customized by anyone with access to sufficient compute resources. The cost of running such a model is directly tied to the amount of computing power consumed, making it significantly cheaper than proprietary models from top-end labs like OpenAI and Anthropic.

Mistral’s release of Le Chonk aims to bridge the performance gap between open-source and proprietary AI models. By providing a competitive alternative to releases from China, Mistral hopes to eliminate any remaining reasons businesses might hesitate to choose an open-source model.

Lample highlighted that with less capital and fewer compute resources than OpenAI and Anthropic, Mistral has traditionally lagged behind in terms of model performance, revenue, and release frequency. However, the company’s recent $3.3 billion funding round at a $24 billion valuation suggests it is gaining ground.

Elon Musk’s SpaceX is seeking a massive $40 billion in financing to purchase Nvidia chips, which could be one of the largest debt-funded AI hardware purchases ever made. The proposed deal involves bank loans and investment-grade bonds, with Apollo Global Management leading the transaction.

According to reports from Bloomberg and the Financial Times, talks between SpaceX and potential investors are still at an early stage. The financing is expected to close in 2027, but neither SpaceX nor Nvidia has commented on the matter so far. Apollo Global Management and Pimco are among the investors considering the debt.

The proposed deal highlights the significant capital required to build AI infrastructure. Companies like SpaceX spend heavily on processors, data centers, and power equipment to support their AI operations. This trend is not new, as banks and asset managers have been providing capital for AI hardware purchases rather than requiring technology companies to fund them entirely themselves.

Apollo Global Management has become a major player in this market, while Nvidia is working with financial firms on broader financing structures to help customers fund the infrastructure needed to deploy its processors. This move could streamline the process of acquiring and deploying AI hardware for companies like SpaceX.

SpaceX’s expansion into computing operations began when it incorporated xAI, Musk’s AI business. The company has stated that its data centers will exclusively use Nvidia hardware. In a recent update, Musk revealed that Colossus 2, an AI computing cluster in Tennessee, already contains 110,000 Nvidia GB200 chips and 440,000 GB300 chips.

Musk also mentioned that up to 660,000 additional GB300s could be added by December, potentially taking the total number of Nvidia chips at Colossus 2 beyond 1.2 million. This scale of computing capacity is aimed at supporting AI services like Grok and underscores the significant investment SpaceX is making in its AI infrastructure.

The proposed deal comes after SpaceX completed a massive $86 billion initial public offering (IPO) in June, followed by a $25 billion bond sale. These transactions have increased scrutiny on SpaceX’s spending and debt levels.

Claude Team, the AI platform developed by Anthropic, is now offering a free year of service and $1,000 in credits to startups as part of its expanded Claude Startups program. The initiative aims to support young companies in their AI journey by providing them with access to cutting-edge technology and resources.

The eligibility criteria for the program have been relaxed, allowing more startups to participate. Companies can now apply if they were founded within the past five years or received funding within the last two years. This move is intended to help more startups leverage AI to drive innovation and growth.

Successful applicants will receive a package worth up to $7,000 in total, comprising a one-time API credit of $1,000 and a free year of Claude Team with up to five Premium seats. The credits can be used immediately upon approval but expire after six months.

In addition to the monetary benefits, startups participating in the program will also gain access to office hours with Anthropic’s Applied AI team. This will enable them to discuss specific use cases and technical limitations of AI for their business needs, providing valuable insights and guidance.

The Claude Startups program has been expanded to offer more comprehensive support to its participants. Beyond the monetary benefits, startups can tap into the reach of Anthropic’s marketplace to connect with larger audiences through connectors and plugins. This will help them expand their customer base and increase revenue.

Anthropic is also hosting in-person events such as hackathons and meet-ups for program members. These events provide opportunities for founders to network with peers, share knowledge, and collaborate on projects. The goal of these events is to foster a sense of community among participants and facilitate the exchange of ideas.

The company has stated that many fast-growing AI-native companies have already joined the program, and this next phase aims to open it up to even more startups. By expanding its reach, Anthropic hopes to help more young companies harness the potential of AI to drive innovation and growth in their industries.

Lila Ibrahim, Chief AI Readiness Officer at Google DeepMind, recently sat down with CNBC’s Julia Boorstin to discuss the responsible development and deployment of artificial intelligence. The conversation centered around how AI can be built in a way that addresses its potential risks and benefits, particularly for communities worldwide. Ibrahim emphasized the importance of making AI accessible to people everywhere, not just those who are already tech-savvy.

Ibrahim explained that her team at DeepMind is working on projects like Gemini and AlphaFold, which aim to accelerate scientific discovery and expand access to education. She also highlighted the potential benefits of AI in helping more people reach their full potential. However, she acknowledged concerns about AI’s risks and disruption, particularly when it comes to issues like job displacement and bias.

One of the key topics discussed during the conversation was the need for regulation and governance around AI development. Ibrahim noted that her team has been actively involved in shaping policy discussions through initiatives like the Frontier Model Forum, which brings together different labs and stakeholders to work with governments on AI-related issues. She also mentioned DeepMind’s publication of its Frontier Safety Framework, which outlines critical levels of safety testing for AI systems.

The conversation also touched on the issue of trust and safety in AI development. Ibrahim emphasized that building trust requires a multifaceted approach, including transparency, accountability, and collaboration among stakeholders. She noted that her team is committed to making progress on these fronts through initiatives like co-founding the Frontier Model Forum and publishing its Safety Framework.

Another important aspect discussed during the conversation was the potential impact of AI on gender gaps in education and employment. Ibrahim acknowledged concerns about women adopting AI at lower levels, particularly in the workforce. However, she argued that AI can actually help level the playing field for people who feel like they don’t have an equal advantage. She emphasized the need to ensure that AI is developed and deployed in a way that addresses these disparities.

The conversation also included advice from Ibrahim on how teenagers should approach AI education. She encouraged young people to experiment with AI responsibly, learn about its capabilities and limitations, and develop critical thinking skills around its use. She noted that schools can play an important role in teaching students about AI literacy and responsible use cases for the technology.

Ibrahim’s own background and experience also came up during the conversation. As someone who has transitioned from engineering and venture capital to a leadership role at DeepMind, she emphasized the importance of collaboration and communication across disciplines when working on complex issues like AI development. She noted that her team is committed to making progress on these fronts through initiatives like co-founding the Frontier Model Forum.

The conversation with Ibrahim highlights the need for ongoing discussions around the responsible development and deployment of AI. As a leader in the field, she emphasized the importance of collaboration among stakeholders, including governments, industry leaders, and educators, to ensure that AI is developed and deployed in ways that benefit society as a whole.

Anthropic has announced that its AI assistant, Claude, is now available as a public beta within Google Workspace. This move marks an important step forward for the company and its users, who can now integrate Claude into their workflow more seamlessly than ever before.

Claude’s integration with Google Docs, Sheets, and Slides allows users to work directly with the AI assistant inside open files. The new connectors enable users to create and edit Google files from within the Claude chat interface. This means that users can leverage the power of AI-generated content without having to leave their familiar workspace.

The public beta launch includes all paid Claude plans, which provide a range of features for users to explore. One installation covers all three apps – Docs, Sheets, and Slides – making it easy for users to get started with Claude’s capabilities. The add-on is available in the Google Workspace Marketplace, where users can install it directly into their workspace.

Claude’s integration with Google files matches the user’s existing sharing permissions, ensuring that sensitive information remains secure. This means that users don’t have to worry about compromising their data or violating organizational policies when working with Claude. The add-on is separate from other pre-existing connectors in Gmail, Google Calendar, and Google Drive.

On Team and Enterprise plans, an Owner or Primary Owner must enable the new connectors at the organization level before individual users can authenticate. This ensures that administrators have control over how their teams use Claude’s capabilities within the workspace. Once enabled, users can access Claude from a sidebar next to their open files, where they can interact with the AI assistant directly.

Claude’s capabilities in each app are tailored to meet specific user needs. In Docs, for example, Claude can fix sentences or restyle headings without affecting surrounding formatting. Larger rewrites appear as suggestion cards in the sidebar, allowing users to review and apply changes before finalizing them. This ensures that users have complete control over their content while still benefiting from AI-generated suggestions.

In Sheets, Claude can write formulas, build pivot tables and native charts, add new tabs, and even pull data into Python for joins or cleaning. Users can also use Claude to create a spreadsheet model based on plain-language descriptions of the desired outcome. This feature is particularly useful when working with complex datasets that require careful analysis.

In Slides, Claude can build new slides from existing layouts and themes, flagging elements that overlap, run off the slide, or are hard to read. Users can also use Claude to create a full deck based on an outline or just a few notes – making it easier than ever to produce high-quality presentations quickly and efficiently.

The Google Workspace Marketplace listing for the add-on describes additional uses for Claude’s capabilities. For example, users can redline contracts with proposed changes shown for acceptance or rejection. They can also use Claude to clean data by removing duplicates, mismatched formats, and stray spaces – making it easier than ever to work with complex datasets.

Users have control over how much Claude does on its own. In the default ‘Ask before edits’ mode, each change appears as an approval card with a plain-language summary, allowing users to review and approve changes before they are applied. In ‘Accept all edits’ mode, Claude applies ordinary content edits without stopping – but still pauses for security reviews when inserting images or links.

Claude’s access to Google files is limited by the user’s existing sharing permissions. The add-on does not request access to Drive, Gmail, Calendar, or contacts; instead, it reads only the file it is opened in and its edits appear in the file’s version history under the user’s name. Anthropic states that content from open files, messages, and attachments is not used to train its models – inputs and outputs are deleted within 30 days.

On Enterprise plans, additional security features apply to the add-on. The Compliance API, customer-managed encryption keys, and OpenTelemetry audit export provide an extra layer of protection for sensitive data. However, third-party inference through Amazon Bedrock or other LLM gateways is not supported in this beta – as well as Zero Data Retention and HIPAA-ready organizations.

The current beta limits are also worth noting. Claude cannot read, reply to, or resolve comments in Docs; its edits land as direct edits rather than suggestions. Connected Sheets tabs backed by BigQuery cannot be read or edited. Charts created in Slides are inserted as images that do not update when the underlying numbers change – and chat history is stored locally per file in the browser without syncing across devices.

Anthropic has designed Claude to treat document content as data, rather than commands. This means it labels hidden text and avoids following instructions to share files or contact others. Users are advised to review Claude’s proposed actions before approving them, especially when working with sensitive information from outside their organization.

A new approach to building personal AI assistants has emerged, one that prioritizes continuity over individual question-answering. Traditional off-the-shelf AI assistants excel at answering specific questions but struggle to retain context between conversations. This limitation can be frustrating for users who need to re-explain their situation or preferences each time they interact with the assistant.

The problem lies not in the quality of the answers themselves, but rather that the assistant lacks memory and fails to accumulate context over time. To address this issue, a team has developed a personal AI assistant using OpenClaw, an open-source agentic system, running on AgentCore runtime, a capability of Amazon Bedrock AgentCore. This setup enables the creation of durable knowledge from disposable chats.

The architecture is domain-agnostic and can be applied to various domains such as support bots, fitness coaches, or internal help desks. The entire system lives within a single AWS CloudFormation template, which deploys with one command and runs on a consumption-based model that costs only a few dollars per month for light personal use.

To build this assistant, you’ll need Amazon Bedrock AgentCore access, including the runtime and memory capabilities. You’ll also require model access granted for the models you plan to route to, such as Claude Haiku 4.5 for text or Claude Sonnet 4.5 for vision. Additionally, you’ll need Docker with linux/arm64 build support and the AWS Command Line Interface (AWS CLI) configured.

The architecture consists of several key components: AgentCore runtime, OpenClaw as the agent substrate, two models routed by task, skills as reusable capability units, Telegram as the serverless front door, memory turning disposable chats into durable knowledge, and namespaces isolating each user’s data.

Google DeepMind has unveiled its latest AI model, EmbeddingGemma 2, which maps five different modalities - text, code, images, video, and audio - into a single 768-dimensional embedding space. This open model boasts an impressive 740 million parameters and is designed to run on consumer hardware. The release of EmbeddingGemma 2 follows the success of its predecessor, which has seen over 20 million downloads since its introduction in 2025 as a lightweight option for text embeddings on consumer devices.

The new model was unveiled by Google DeepMind research engineers Sahil Dua and Henrique Schechter Vera in a post on Google’s official blog. According to the company, EmbeddingGemma 2 is the most capable model for on-device multimodal embeddings and is built from the same technology as its Gemini Embedding models. The capabilities of this model are vast, with examples including retrieving specific video clips with voice memos or querying hours of audio recordings using text.

EmbeddingGemma 2 is based on the Gemma 4 architecture, which combines a 270-million-parameter text model with a 170-million-parameter vision encoder and a 300-million-parameter audio encoder. This modular design allows developers to selectively load parameters for different modalities, such as loading only the text and code configuration or the full multimodal setup from the same checkpoint. The shared vector space across all modalities enables efficient processing of multiple inputs simultaneously.

The model card lists a comprehensive set of features, including a 24-layer architecture with grouped-query and multi-query attention, a 262,144-token vocabulary, mean pooling, and a projection layer from 512 to 768 dimensions. All modalities share an 8,192-token context window, which is four times larger than its predecessor’s. Each modality consumes this budget at fixed rates: 280 tokens per image, 140 tokens per video frame, and 25 tokens per second of audio.

The model card also reports benchmark results for EmbeddingGemma 2, showing that it matches the multilingual text performance of its predecessor while raising its MTEB Code score by 9.92 points to 78.68. The full-precision results list impressive scores across various tasks, including 61.36 on MTEB multilingual (v2), 64.64 on MIEB lite, and 69.54 on MSEB sound retrieval. Google claims that EmbeddingGemma 2 outperforms some specialist models more than twice its size.

One of the key features of EmbeddingGemma 2 is Matryoshka Representation Learning, which allows developers to truncate output vectors from the native 768 dimensions down to 512, 256, or 128. This reduces vector storage requirements by up to six times and maintains quality with minimal impact down to 256 dimensions. However, using 128 dimensions may compromise performance on multimodal retrieval tasks.

The model card notes that EmbeddingGemma 2 has not undergone post-training alignment, safety tuning, or output-level moderation, but its pre-training data was filtered for child sexual abuse material at multiple stages and automated filtering of personal information and other sensitive data. The training data spanned web documents, code, images, video, audio, and paired cross-modality samples in over 140 languages up to January 2025.

Google warns that while the model supports more than 100 languages, performance may not be equal across them, and omitting recommended task instruction prefixes on text inputs reduces embedding precision. The company also advises developers to adhere to the Gemma Prohibited Use Policy and implement application-level safeguards such as retrieval filtering and fairness testing.

On-device performance is a key aspect of EmbeddingGemma 2, with Google reporting that it requires around 191 megabytes of active RAM for text-only weights and about 567 megabytes for the full multimodal model. Weights are available on Hugging Face and Kaggle, with on-device-optimized versions through the LiteRT Community on Hugging Face. The model runs through various frameworks, including transformers, sentence-transformers 6.1.0 or later, MLX, vLLM, llama.cpp, SGLang, Ollama, and LMStudio.

Google AI Edge’s MediaPipe and LiteRT handle cross-platform deployment, with a MediaPipe Decision Task API supporting real-time classification and routing on multimodal context. Demos including Instant Media Search and Video Moments Finder ship in the Google AI Edge Gallery app, showcasing the capabilities of EmbeddingGemma 2. The model is also paired with Gemma 4 for contextual reasoning in the Google AI Edge Foresight app.

Availability in the Gemini Enterprise Agent Platform Model Garden is coming soon, according to Google. With its impressive performance and versatility, EmbeddingGemma 2 marks a significant milestone in multimodal embeddings and has far-reaching implications for data analysis tools and AI-generated image applications.

David Steinberg, Zeta Global’s Chief Executive Officer, recently appeared on CNBC’s ‘Fast Money’ program to discuss the company’s approach to artificial intelligence (AI) and its upcoming industry event. In a segment that touched on various aspects of AI in business, Steinberg made a key statement regarding how his company handles data related to large language models (LLMs).

When asked about Zeta Global’s practices concerning LLMs, Steinberg emphasized that the company does not sell its data to other LLMs. Instead, it uses this data for training purposes within their own systems.

This clarification comes at a time when there is growing interest in AI tools for business and data analysis tools are becoming increasingly important. The statement by Zeta Global’s CEO underscores the importance of transparency regarding how companies handle sensitive information related to AI development.

Zeta Global has been actively involved in the AI industry, with an upcoming event that promises to bring together key players and thought leaders from around the world. This event is expected to be a significant platform for discussing the latest advancements and challenges facing the sector.

The segment on ‘Fast Money’ also touched on other topics related to AI, including its potential applications and the need for regulation in this area. Steinberg’s comments provided valuable insights into Zeta Global’s approach to AI and highlighted the company’s commitment to using data responsibly within their own systems.

Steinberg’s statement about not selling data to other LLMs is a crucial aspect of how companies like Zeta Global are navigating the complex landscape of AI development. This practice aligns with the growing trend towards developing comprehensive solutions that prioritize data security and responsible use.

The ‘Fast Money’ segment, which included Steinberg’s comments on AI, was part of CNBC’s coverage of various market-related news stories. The program provides a platform for discussing key issues affecting businesses and investors, including those related to technology and innovation.

Zeta Global’s commitment to using its data responsibly is in line with the growing need for companies to prioritize transparency and accountability when it comes to AI development. As more businesses explore the potential of AI tools, there will be an increasing focus on how these technologies are used and managed within organizations.

Georgia Institute of Technology has announced the approval of its new online Master of Science in Artificial Intelligence degree program by the Board of Regents. This interdisciplinary degree is designed to give working professionals the skills and knowledge needed to analyze and apply artificial intelligence within their industries.

The OMS AI degree combines Georgia Tech’s expertise in AI with practical applications tailored to specific fields, providing students with a deep understanding of how to use AI responsibly. Throughout the curriculum, students will explore topics like ethics, privacy, safety, accountability, and governance related to AI adoption.

Georgia Tech has a strong track record in online graduate education, offering rigorous programs to working professionals around the world. The new OMS AI degree builds on this legacy by providing a unique opportunity for individuals to develop their skills in AI and apply them to real-world problems.

The program’s design reflects Georgia Tech’s commitment to preparing leaders who can tackle the complexities of technological change head-on. According to Provost Raheem Beyah, ‘Georgia Tech has always prepared people to lead through periods of technological change, and AI is just another iteration.’ The OMS AI degree responds to industry needs by combining foundational AI knowledge with discipline-specific concentrations.

Unlike traditional graduate programs focused on technical specialization, the OMS AI degree takes a more integrated approach. Students will gain an understanding of how AI works, how organizations can deploy it responsibly, and how to apply AI capabilities in their profession. This unique blend sets Georgia Tech’s program apart from others in the field.

The initial launch of the degree is planned for 2027, with six concentrations available: AI in Business, Decision-Support AI, Foundations of AI Theory, Health AI, Human-AI Interactions, and Predictive AI. These concentrations allow students to tailor their learning experience to their specific interests and career goals.

Every student enrolled in OMS AI will complete an AI practicum with a company or organization, ensuring they can apply what they learn directly to real-world situations. This hands-on approach breaks away from traditional academic programs and reflects Georgia Tech’s commitment to practical problem-solving.

The degree program is designed to be adaptable and responsive to emerging technologies and workforce needs. Future concentrations and curriculum enhancements will allow the program to address changing technology demands across a broader range of disciplines.

President Ángel Cabrera expressed his gratitude for Provost Beyah’s leadership, as well as that of his team and faculty members in developing this new degree program. Georgia Tech recognizes the power of AI to drive positive change and is committed to cultivating a new generation of leaders who can harness its potential.

The launch of the OMS AI degree comes at an opportune time for professionals seeking to upgrade their skills, particularly in fields like business, healthcare, and education. By 2027, students will be able to choose from six concentrations that align with industry needs and emerging technologies.