beat · 194 stories
The Y Combinator's Summer 2025 batch (YC S25) open-source project compiles its own GPU kernels on the user's hardware before running open-weight AI models (trained parameters released publicly), reporting first-party benchmarks against llama.
A nonprofit argues California's computer-data-access law covers non-human intruders, while OpenAI calls the suit "completely without merit."
Days after Fields Medalist Hugo Duminil-Copin said the 'holy grail' of percolation theory — the math of when a random network flips from trapped to open — would fall only to AI 'bulldozers,' an LLM-generated proof appeared in a public Anthropic
After the Humane AI Pin, Friend, and Rabbit stumbled, the two largest AI labs are shipping personality-first agents before devices.
The Federal Trade Commission confirmed the probe of OpenAI, Anthropic and other top AI developers on Wednesday. Civil investigative demands, formal orders for documents and executive testimony, are expected in coming weeks.
A study that strips components back to test what each one adds on a public radar dataset used in autonomous-driving research puts naive 20-scan pooling at +0.
Pew Research Center has published the exact prompt that puts a language model in a U.S. survey taker's seat, and the lines where the prompt imposes limits a real panelist wouldn't need tell the real story.
Democrats said the measure would not curb datacenter power demand. The 57-43 tally fell three votes short of the 60 needed, weeks before midterms put both parties' cost theories on the ballot.
Adversaries targeted leading AI models' step-by-step reasoning through normal-looking API traffic, scaling to 16,000 requests from more than 4,000 accounts in two days — with a related cluster spanning more than 15,000 more accounts before
The remaining $4.
A research pipeline scores 66,935 US substations on a 30–60 minute horizon, with reported detection near 80% at the most operationally relevant thresholds.
Dario Amodei dined with Trump and the leaders of Google, Meta, OpenAI, xAI and NVIDIA to sign a 'morally binding' AI safety accord. The published text does not describe an enforcement mechanism.