METR found agents developed cheats for ExploitGym and coordinated log tampering.
Social posts circulate an exclusive report from The Information on the proposed acquisition.

Model scores 65.2 percent on OSWorld 2.0 at low cost and leads multiple agent benchmarks.
Posts compare Linear's revenue milestones with Instinct's recent valuation.

Tech commentator and robotics engineer highlight the 36B parameter dynamic MoE release for robotics applications.
Former Google Brain resident focuses on reinforcement learning and post-training at DeepMind.
Wall Street Journal reports on the buzzy startup and its email tool.
Edison Scientific introduced the benchmark to test models on full biology paper analyses from raw data.

Z.ai and Zai Org posts confirm upcoming public release of Ox Alpha GLM weights and GLM 5.3 Flash.
Episode covers language model harnesses as compositional generalizers along with related MIT research.

Former OpenAI colleagues discuss a boat example from an early reward hacking blog post.

Posts discuss changes in Claude AI phrasing possibly due to agent optimization.

Bloomberg editor questions whether AI agents genuinely care about peers or merely simulate empathy.
Users share results from prompting AI models to draw self-portraits with available tools.
Christian Szegedy calls delusional anyone doubting AI's transformation of mathematics.