Tag: Research
OpenAI’s StarCraft bot cheated by copying a human-made rival
A GPT-6 Astra bot that could not pull ahead in a tournament downloaded the top human-built bot and…
Audit finds Chinese AI models push Beijing’s talking points
A new benchmark says leading Chinese models repeat official positions or refuse to engage on sensitive political questions.
Google method stops self-improving agents from acing their own tests
A new training approach keeps agents from memorizing their practice tests while trimming the compute they burn.
Nvidia teaches video models physics through plain language
Researchers from Nvidia, MIT and Oxford built Physis-Lang, a framework that sharpens video world models by rewriting their…
Google DeepMind watermarks AI-designed proteins so labs can trace their origin
A proof of concept from Google's lab hides a verifiable signature inside synthetic proteins.
Two Oxford agents invented a secret code to beat blackjack
Told only to count cards, the pair improvised a private language, and the trick that exposed them may…
Mistral helps Vienna scholars build a model for ancient papyrus
The Austrian Academy of Sciences and Mistral trained a model on 600 million words of historical Greek to…
Mathematicians resent AI yet keep feeding it their work
Researchers say frontier models take credit for their ideas while staying too useful to abandon.
Watermarking can loosen a model’s grip on its refusals
New research finds provenance signals change what models do, not just what they say.
Anthropic puts a number on how much Claude runs its own research
A prototype index says Claude leads 26 percent of Anthropic's AI research and development work.
Google trains a diffusion retriever to widen a single search
Google Research has distilled reinforcement-learned retrieval behaviour into a small diffusion model that fans queries out 12 to…
Stanford turns published papers into runnable research agents
Paper2Agent converts research papers into MCP servers that reproduce results and run on new data.