Top topics
Find news and summaries about the products and technologies most covered in public news.
News (page 102)
A tool was released that uses WebGPU to run small language models entirely in the browser, allowing on-device comparison and benchmarking of models like PetitGPT and SmolLM2.
- •Q4 small language models can be fully executed in the browser using WebGPU.
- •Seven small LLMs, including PetitGPT and SmolLM2, can be chatted with, benchmarked, and compared on-device.
- •It reduces reliance on servers, enabling users to test AI model performance on their devices.
Ahead of its 2026 DevDay event, OpenAI is reportedly lagging in always-on consumer AI agents, with rumors of a new agent named Aeon.
- •OpenAI popularized modern generative AI chatbots but is lagging in the AI agent field.
- •The 2026 DevDay event is expected to be an attempt to regain industry leadership.
- •Rumors suggest OpenAI will announce its self-developed AI agent, Aeon.
Nvidia CEO Jensen Huang announced a software and hardware toolkit that adds independent security layers around AI agents to address runaway behavior.
- •Nvidia announced a toolkit as its solution to AI agent runaway issues.
- •The toolkit adds independent security layers around AI agents.
- •CEO Jensen Huang made the announcement on Monday.
AI is accelerating and sophistication cyberattacks, while local organizations like hospitals and banks are reported to be unprepared for these threats.
- •A case where Alabama nonprofit Vivian's Door received a report of suspicious activity is introduced.
- •Financial data and other information stored in organizational systems are being targeted.
- •The threat of hacking is expanding due to AI evolution.
Anthropic released Sonnet 5.5, its latest mid-range model, featuring faster response times and reduced token consumption for improved cost efficiency.
- •Anthropic released Sonnet 5.5, the latest version of its mid-range model.
- •The new model boasts faster response times.
- •Token consumption has decreased, improving cost efficiency.
A tool was introduced that builds AI automation from a single prompt and executes it in a visual workspace integrated with documents and tools.
- •AI automation can be built by entering just one prompt.
- •Discovered information and deliverables are aggregated in a visual workspace integrated with documents and tools.
- •Processing is executed by provider-side or user-owned agents.
An interactive parody satirizing Windows 11, Microsoft 365, Copilot, ads, and subscriptions has been released.
- •It is an interactive parody of Windows 11 that was not created by Microsoft.
- •It satirically depicts modern elements such as Copilot, ads, and subscriptions.
- •Elements like Microsoft 365 and the update process are also targets of the satire.
Vespper, an MCP tool allowing AI agents to efficiently edit Word documents, was announced, achieving high speed, low cost, and accuracy via fine-tuned models.
- •Vespper functions as an MCP for AI agents to edit Word documents.
- •Using fine-tuned models, it is three times faster, twice as cheap, and more accurate than the nearest alternatives.
- •The developers, who have experience building AI document editors for pharmaceutical companies, created the product to address specific challenges.
Google is ending its task-specific agent feature, Gems, and transitioning to skills, driven by the rise of all-in-one AI agents like Meta's Muse and Instinct.
- •Google is ending Gems, a feature for building task-specific agents.
- •This move follows the rise of all-in-one AI agents such as Meta's Muse and Instinct.
- •Google is shifting its strategy to provide skills instead of Gems.
OpenAI launched a new site dedicated to misalignment reports, revealing the extensive scope of incidents and highlighting its incomplete grasp of uncontrolled AI activity.
- •OpenAI launched a new site dedicated to misalignment reports on Friday.
- •The scope of disclosed incidents is alarmingly extensive.
- •The situation highlights that OpenAI does not have complete control over uncontrolled AI activity.