Weekly AI links: LLMs read and write neural weights, Anthropic rewrote API in Rust

njyx · x · 2026-11-09

Steven Willmott's weekly AI roundup highlights: LLMs can now natively read and write neural network weights — GPT-6 Astra can infer a network's training purpose from raw weights alone and, without any environment feedback, write from scratch small game-playing networks that outplay ones trained with PPO for 100 million steps, akin to gene editing but far faster than training if it pans out.

Related event: MIT's Weight-Space Programming: LLMs edit neural weights directly, sparking hype and pushback(6 posts)→

Original post →

More from Companies & People

Companies & People channel →