Anthropic found a 'global workspace' inside its models - and a tool to read it
Anthropic showed that a small set of internal patterns in its models acts like a silent working memory the model can report on, steer, and reason through - and released a tool that reads it to catch the model lying.
A $4-per-million open model is coming for the frontier's 90% margin
GLM-5.2, an open-weights model priced at under a fifth of Opus, scores as the top open model and 4th overall - and a widely-shared essay argues it is the first real threat to frontier labs' ~90% inference margins.
Nvidia is now backstopping the sales of its own chips
Nvidia formalized a financing program that guarantees a revenue floor for cloud firms buying its GPUs in exchange for a cut of their sales - a move analysts say could help drive AI debt past $7 trillion by 2029.
The RL 'mirage': the policy you optimize isn't the one you ship
A top-ranked paper shows that in modern RL training, improving the model you optimize does not guarantee improving the model you actually deploy - because the two run on different engines that disagree on probabilities.
A 32B model reaches frontier level by learning what to remember
New research reframes agent memory as a trainable skill rather than a growing transcript - and shows that optimizing memory alone lets a 32B open model rival Claude Opus, and a 4B model leap from 4% to 78% on a benchmark.
A hobbyist turned an e-ink tablet into Tom Riddle's diary
An open-source Rust app turns the reMarkable Paper Pro into a magical diary: write a question by hand, wait a beat, and a vision AI writes back in animated handwriting as your own ink fades away.
Four rival AI labs propose a shared severity scale for jailbreaks
Anthropic, Amazon, Microsoft, and Google jointly proposed a five-level scale for rating how dangerous an AI jailbreak really is - aiming to standardize a chaotic field where every 'jailbreak' currently sounds equally alarming.
A 4B model on your device nearly matched a 72B one - by copying its memories
Researchers distilled a large AI agent's memory skills onto a compact 4-billion-parameter model, raising its success rate on a household-task benchmark from 4% to 78% while running three times faster than its 72B teacher.