Blogger reproduces Qwen 3.8 27B weight-interpretation experiment on a laptop
Understand this story
Simon Willison reproduced the paper's "identifying the game from weights" experiment, saying that a locally run Qwen 3.8 27B could infer a neural network's training objective from raw weights alone, performing similarly to GPT-6 Astra. When loading Qwen 3.8 27B in LM Studio, a long context window is required; the default 8192 context throws an error. He ran it via the llm-lmstudio plugin. The model guessed Pac-Man in 46 minutes 12 seconds using 43,506 reasoning tokens, guessed Breakout in 52 minutes 25 seconds with 88% confidence, but failed to identify Freeway. The report also noted that the Breakout prompt can be run against Kimi K3.1 Pro on OpenRouter via the llm-openrouter plugin.
AI overview based on source reports · Updated 1 days ago
Coverage timeline
Follow the coverage to explore different perspectives.
- Simon WillisonQwen 3.8 27B on my laptop can interpret model weights, too
Simon Willison tested the "Sparks of Weight-Space Programming" paper's game-identification prompts, which ask a model to name which of four Atari-style games a policy network was trained on using only its raw weights and no tools. Running Qwen 3.8 27B locally via LM Studio, he got correct answers for Pac-Man (46m 12s, 43,506 reasoning tokens), Pong (1h 4m 54s) and Breakout (52m 25s), but it failed on Freeway. He notes critics like Matthew Honnibal and Sean Goedecke dispute the paper's framing, and he focuses on the no-tools question rather than claims that source code is obsolete.
Activity for this story
- Current comparable activity
- 3
- Peak comparable activity
- 10Nov 4 02:00
- Change in the past 24 hours
- -51%
Trends compare the same fully observed participants, which may cover fewer participants than current activity. Move the pointer or tap the chart to inspect hourly activity; use the left and right arrow keys with a keyboard.