Independent AI news, with every claim linked to its primary source
Independent AI news, with every claim linked to its primary source
A study of 171,264 donated chatbot conversations and 1,000 replayed prompts found four platforms disagree almost completely about when a question needs the web, and that searching more did not produce better answers.
Read More
ScienceIDE converts 27 scientific codebases into 64 executable environments and 2,812 graded tasks. On the 85-task hard subset the best agent scored 67.1%.
Read More
Anthropic has rebuilt projects in Claude Code so that one request fans out into a team of cloud sessions directed
Read More
A three-person team at the security startup Hacktron used Anthropic’s Claude to break into OpenAI, and they did it in
Read More
Moonshot AI launched a financial industry package for its Kimi models on Thursday, wiring the assistant into more than 10
Read More
AI technology is a stack of six layers, from silicon to agents. Each one has its own vendors, its own unit of measurement and its own bottleneck. We walk all six using NVIDIA and Google hardware specs, Epoch AI trend data, two vendor price lists, three papers and two protocol specifications, and name where the top of the stack still falls down.
Read More
Jensen Huang told Dreamforce and CNBC that AI safety is an engineering problem, so new laws are unnecessary. Mark Zuckerberg backed him the same day.
Read More
Adoption of AI apps stalled at 64% of US adults this year while spending tripled. Here is what App Store, Sensor Tower, Pew and Menlo data show about which ones survive past the first month, and what they cost.
Read More
Evvy said on Tuesday it has closed an oversubscribed $40 million Series B led by Catalio Capital Management. The women’s
Read More
Broadcom CEO Hock Tan told CNBC on Monday that the chipmaker’s AI revenue targets haven’t moved. Anthropic CEO Dario Amodei
Read More
Google shipped an early access MCP server for Google Home, letting outside AI agents read device history and issue commands for US Premium Advanced subscribers.
Read More
A ReAct agent on GPT-4.1 passed all five runs of an AppWorld task just 53.0% of the time. IBM Research built a diagnostic that raises that to 69.0% without costing average accuracy.
Read More