Qwen 3.8 27B on my laptop can interpret model weights, too
Web page · https://simonwillison.net/2026/Nov/3/qwen-can-interpret-model-weights
First seen by Oriel on .
AI summary
Simon Willison tests whether LLMs can identify games from neural-network weights without tools, finding that Qwen 3.8 27B on a laptop correctly recognizes three of four game-playing networks.
- Qwen identified Pac-Man, Pong, and Breakout from their weights, but misclassified the less familiar Freeway network as Pac-Man-like.
- The tests used self-contained prompts containing floating-point weights for reinforcement-learning policies trained on simplified grid versions of four classic games.
- Qwen’s correct answers took 46 minutes to over an hour and used tens of thousands of reasoning tokens; the default 8192-token context failed.
- The author distinguishes interpreting weights without tools from modifying models using code, and cautions that the findings do not justify claims that source code is obsolete.
#llms #neural networks #model weights #reinforcement learning
Generated by gpt-5.6-sol. Check the original for details.
Metric trends
Last 30 days · Up to 120 recorded samples per source and metric, with reposts kept separate · First and latest retained
Points
Score reported by the source.
- Hacker News: 713, latest observation ; +658 over observed period since
Comments
Comment count reported by the source.
- Hacker News: 206, latest observation ; +200 over observed period since
Aliases
| Source | External ID | URL |
|---|---|---|
| Hacker News | 50355251 | https://simonwillison.net/2026/Nov/3/qwen-can-interpret-model-weights/ Read discussion |