MIT found brain-like modules inside six large language models
MIT researchers localized the neurons behind 46 reasoning tasks in six large language models and found that tasks sharing a brain network in humans share neurons in the models, with 4.3 times more overlap within a cognitive domain than across domains.
A 9B model writes agent upgrades as good as Claude Opus 4.6
A 17-author study separates the ability to improve an AI agent's scaffolding from the ability to benefit from the improvement, and finds that a 9-billion-parameter model produces upgrades yielding gains comparable to Claude Opus 4.6.
Agent skill libraries now need a librarian, not a folder
SkillsVote profiles a million-scale corpus of open-source agent skills and admits new ones only through evidence-gated updates after execution, arguing that indiscriminate additions to a skill library pollute every future run's context.
Compressed memory stretched a 7,000-token model to 1.75 million
A framework that chunks long documents, compresses each chunk into memory blocks and gates which blocks the model reads extrapolated from 7,000 tokens of training context to 1.75 million at inference, with half the peak GPU memory of a leading baseline.
EXO keeps an agent's memory outside the code the agent rewrites
The EXO agent runtime splits a self-modifying agent into a disposable policy layer and a durable state layer, so an agent can rewrite its own prompts, tools and executor code without being able to damage its own event log, secrets or history.
OpenAI hands its offensive cyber models to sixteen security firms
OpenAI expanded its Daybreak Cyber Partner Program to sixteen named companies including Accenture, IBM, Cisco, CrowdStrike and Cloudflare, letting them embed its frontier cyber models in their own products while keeping model access away from end customers.
A fired xAI engineer says he was cut days before presenting safety findings
A wrongful-termination complaint filed in Santa Clara County alleges an early xAI engineer was fired shortly before presenting AI-safety findings to leadership, and it sits against a verified record of a Canadian regulator ruling that Grok's image tool launched without proper safeguards.
Qwen passed one billion downloads, not three billion
Alibaba Cloud's own announcement says its Qwen models crossed 1 billion cumulative Hugging Face downloads at about 1.1 million a day with 200,000 derivative models, passing Meta's Llama, contradicting a widely repeated claim of 3 billion downloads past both Meta and Google.
The open world model ships inference and keeps the training code
AlayaWorld released inference code and pretrained weights for an interactive world model with long-horizon memory, but the training code is still an unchecked box, the license is a community license, and running it requires a gated Google model plus a ByteDance depth model.
One layer creates the giant activations behind attention sinks
Researchers identified a single layer, consistent across model families, where the outsized activations that produce attention sinks first appear, and showed that loosening that token's rigidity improves instruction following and math reasoning without retraining.