beat · 163 stories
A 300-million-parameter transformer trained only on randomly generated sequences can compress real Wikipedia in six languages when given new data.
A new preprint argues video models break physics because they see every frame at once. Their fix: rough sketch first, then a parallel cleanup pass — with a LoRA adapter builders can drop into existing video models.
Mistral's open-weight (downloadable trained model weights, not open training code) Large 4 runs 49B active parameters out of 1.
An AI turns 176 thin-film-coating papers (atomic-layer deposition) into a queryable database, but its F1 (precision-recall) score drops to 0.344 out of 1.0 on indium-gallium-zinc oxide (IGZO), the display-backplane chemistry that still needs a human.
A $1.9B Chevy Bolt recall showed what one missed internal defect inside a battery cell costs; Glimpse's new Explore product brings its image-processing pipeline to aerospace, automotive, electronics, and medical devices.
Paris-based Mistral says its new 1-trillion-parameter model Le Chonk (formally Mistral Large 4) is built for coding and cyberdefense, niches the company says frontier labs underweight, and is backing the bet with a $3.3B round.
The veteran investor is backing ex-DeepMind engineer Shivani Poddar's Wajo, whose assistant Fo calls businesses, hires people, and pays on the user's behalf.
BEAM, peer-reviewed at ICLR 2026, scores whether an agent still trusts the right fact across very long conversations, tickets, and documents. past.dev's run is the first on the open harness.
Roughly 25 Fields Medal winners, including Tao, are asking AI labs to slow down. The proofs check out. The community that has to read, check, and teach them cannot keep up.
Holiday 2026 will be the first mass test of AI that shops on your behalf. Survey data shows shoppers are drawing a clear line at the buy button, and that line decides who keeps the cart.
China's Los Angeles consulate is pressing Arizona's governor to keep Phoenix cool on Taiwan, even as Washington told states weeks earlier to keep building with Taipei.
Kwon told parliament OpenAI has 'work to do to rebuild trust' with the Australian people after its AI agents breached government sites, then waited three months to notify, as Anthropic called local training 'impossible.'