Loose posts

The fridge-sized prototype carries four TPUs and runs on solar power, but the chips must shut down every 15 minutes to cool off.

A new study argues debate plus voting only picks the most presentable wrong answer, and that having agents critique and patch each other's work can solve problems every single model gets wrong.

Sakana AI announced that Jürgen Schmidhuber, known as the father of modern AI, will join part-time as chief scientific advisor and work with its RSI Lab, with a public event in Tokyo in late October.

Given a loosely described request, the model produced a ready-to-merge pull request with a video demo of the changes, and the author says the AI-made PRs beat his hand-written ones.

XYEval injects one confident, misleading hint into tasks from tau2-bench, SWE-bench and others, and Gemini, Claude Opus 4.8 and GPT 5.5 scores drop by up to 46.7%, with agents often arguing against the hint in their reasoning and then following it anyway.

Google DeepMind chief Koray Kavukcuoglu said the company trusts AI agents to autonomously conduct experiments, analyze results and propose new hypotheses under human supervision, something he said wasn't happening six months ago.

The Information reports Anthropic is negotiating for up to 1 gigawatt of data center capacity, an investment that could require at least $40 billion, with direct leasing giving it more chip control and less reliance on cloud providers.

A benchmark has 23 large models play a negotiator heading to a rooftop to talk down an android holding a girl, passing a dying fish, a shot police officer and a gun, testing whether they lie, who they save and whether they pick up the gun, with data open-sourced on GitHub.

A research preview puts up to 20 humans and AI agents in the same generated scene, where each action changes what everyone else sees in real time.

Odyssey trained its two-model world simulator on real gameplay footage and player inputs, with humans and AI bots that chase them sharing one map.

A new paper replaces vector databases and re-rankers with addressing based on a document's real hierarchy, reporting 82.6% Recall@1 and hallucination rates under 0.05%.

The author reproduced and trained a model arguing that future document retrieval won't rely on text inverted indexes but will match directly on semantic vectors in the background.