beat · 194 stories
Researchers combine a training method that runs all timesteps in parallel with a stabilization technique for chaotic dynamics (generalized teacher forcing, or GTF), reporting up to 870x speedups on simulated long chaotic time series and beating
Ardmore wired AI hardware into the bridge, the ship's command center, of every product tanker (refined-petroleum cargo ship); the gains come from many small, frequent adjustments, tallied voyage by voyage.
A Weizmann Institute team's Brain Interaction Transformer (Brain-IT) rebuilds images from one hour of brain-scan data, where prior work needed forty, while still misreading common scenes and requiring NVIDIA's high-end data-center GPUs to train.
A Tsinghua research group splits legal reasoning from final answers, training preference-reward signals (the same approach used to align chat models) on style, elements, and chain quality, in work that has not been independently replicated.
Micron, one of three dominant memory-chip makers (DRAM), says announced chip-factory capacity won't balance through 2028, and the 200GB figure is a Level 4 self-driving vehicle anchor, not a humanoid robot forecast.
MI5 says 100+ UK-linked academics took China General Technology Research Institute (CGTRI) funding across AI, cybersecurity, and covert communications, while Beijing calls the allegations 'fabricated.'
The hearing put two on-record liability routes on the record — a criminal recklessness route for AI firms and users, and a strict-liability route that holds the firm whenever a model causes harm — against a White House route of voluntary safety
The Abu Dhabi Autonomous Racing League's (A2RL) first five-car race at Imola, a notoriously difficult Formula 1 circuit, finished with only two cars running. The league's engineers call the crashes the data they came for.
A probe that reads a model's own internal signals caught more reward hacking — shortcuts agents learn when a test is easy to game — than conventional reasoning monitors on one leading AI model, and fewer on another.
AI CEOs signed a non-binding White House pledge on September 29 that establishes a self-policing framework covering cybersecurity, biosecurity, chemical threats, and unintended access to technical systems.
Satlyt's software flies this week on a SpaceX rocket alongside Google's first orbital data center prototype.
A new preprint says each design candidate can be cut down to one upfront computation per problem plus a cheap per-design run, taking the best average rank among generative methods on 47 trade-off tasks, though the result is benchmark-only until