OpenAI CEO Sam Altman warned on Alex Heath's Sources podcast that he is seeing 'the first signs of what feels to me like unsustainable silliness' — ...

The paper's move is a small pivot in vocabulary that carries a bigger technical claim: streaming video memory should stop being a bank you query and ...

A new reward model from Fudan University and Tencent Hunyuan beats GPT-5.5 at judging camera-conditioned world-model video, and the margins are ...

A new paper argues personalized assistants should refuse some requests not because the requests are forbidden, but because the user's own ...

Compile by training reaches 83.6% semantic accuracy on FuzzyBench-Hard, a subset where the earlier Program-as-Weights fast compiler produced no ...

Researchers from Fudan, Tencent Hunyuan, SJTU and Shanghai AI Lab publish WorldReward, a chunk-level VLM reward that jointly scores ...

Gimlet Labs, which builds software that splits AI inference workloads across different chip architectures (GPUs and SRAM-centric silicon), raised $300M ...

A new arXiv preprint by Zixuan Fu, Bingxiang He and eleven co-authors argues that on-policy distillation of large language models can be pushed to most ...

Puffin-World, a new unified multimodal model from Nanyang Technological University's S-Lab and collaborators including the University of Michigan ...

A paper posted to arxiv this week proposes training terminal agents against environments that grow harder "generation by generation" instead of ...

OpenAI has pledged $1 billion in subsidized access to its Daybreak cyber models, training and technical support for organizations defending water, ...

OpenEvidence launched three production medical AI models for verified clinicians: Osler (~5s responses), Sackett (~30s), and Snow (~5min for ...