Articles
An unreleased Claude build closed a proof that stumped theorists since 2008
A PSI research system running non-public Anthropic models produced a complete proof of the Aaronson-Ambainis conjecture in three days.
Editor's Picks
Manus raises over $500M to rebuild its agent alone
The Chinese agent developer lands a Boyu Capital-led round at a reported $4B valuation after…
Agentic AI Foundation takes on four projects and reaches ten total
OpenAI, Anthropic, Hugging Face and Alibaba each handed over a component of the agent stack.
TypeSafe banks $870M at a $7.5B valuation weeks after Jev launch
TypeSafe, maker of the non-text Jev model, has raised $870M at a $7.5B valuation led…
Anthropic rewrites its usage rules to ban abusive chats and election meddling
Anthropic's updated usage policy outlaws election interference, weapons software, surveillance and prolonged verbal abuse of…
Latest Articles
Hugging Face teaches a chat model to read raw genetic code
Carbon-X adds DNA tokens to a 14B text model so researchers can paste a sequence and ask about it in…
Jack Clark pulls his self-improving AI forecast forward a year
A benchmark measuring research taste shows capability doubling every three months, and the Anthropic co-founder noticed.
Nvidia cuts robot video labeling to cents with three small models
Flora splits annotation into segments, captions and rewards, and Nvidia says it already labels 200,000 hours a day.
Meta commits large model training to Google chips through a new backend
TorchTPU swaps a brittle bridge for a dispatcher hook, letting PyTorch models run on TPUs with one device string.
Microsoft lets a thousand agents organize themselves on one job
The Agensh harness drops the central orchestrator and lets workers claim their own share of a coding task.
AI lifts junior patent attorneys then lets the gain fade
A three-month trial across eleven law firms found drafting improved with AI, but only senior lawyers kept the benefit once…
A study finds agent harnesses mostly buy you tokens
Two researchers held the model constant across three coding harnesses and found less than two points of difference on 447…
Anthropic opens the sandbox it uses to cage Claude Code
The tool leans on macOS Seatbelt and Linux bubblewrap to wall off files and networks without shipping a container.
