Desk
Papers, model cards, and lab research notes.
A small model trained alongside its own research harness just edged a frontier system on a finance benchmark for a fraction of the price, and the recipe matters more than the scoreboard.
Microsoft's latest research roundup is quietly about one thing: making agents cheaper, more honest about their failures, and useful outside the demo, from codebases to classrooms to monsoon forecasts.
A dementia research consortium is pointing AI at the data that never gets published, and that may matter more than any bigger model.
Retell's growth from 30 million to 100 million monthly calls is a real signal that voice agents are leaving the demo stage, but the numbers come from a vendor showcase.
Check Point cutting vulnerability research from hours to minutes is the defensive case for frontier cyber models, and it raises the bar for keeping humans in charge.
A Convovo study forced two models’ APIs to choose in nearly 500 races. Grok chose Republicans 92% of the time and GPT chose Democrats 97% of the time. What they searched for and whom they cited is the more important finding.
OpenAI is widening its circle of government safety partners to Poland, and the most interesting part is the router bugs Polish defenders already found with its models.
Microsoft's diagnostic AI beat generalist doctors on hard cases in a peer-reviewed simulation, but the real test starts now in a Boston emergency department.
Microsoft Research wants AI agents to make charts people can click, filter and explore, and it's tackling that by handing the hard part to a compiler instead of the model.
An investment research firm and Anthropic ran a genuinely interesting autonomous protein-design experiment, then sold it as drug discovery. The science got the easy part; the marketing skipped the hard part.
Uber is handing an agent to more than 30,000 employees, a big real-world test of whether agentic AI for office work delivers outcomes or just activity.
Recursive's first model is small, free and runs on a laptop, but the real release is a proof of concept for AI that helps build the next AI.