Qwen 3.8 27B on my laptop can interpret model weights, too
Summary
Simon Willison discusses weight-space programming and reports that Qwen 3.8 27B running locally on a laptop can interpret and even modify model weights. The article covers experiments on simple gridworld games (Breakout, Pong, Pac-Man) with prompts and reasoning traces, noting Breakout, Pong, and Pac-Man succeed while Freeway fails. It highlights implications for on-device LLM capabilities and raises questions about the broader potential and limits of weight-editing by LLMs.