Simon Willison Tests Whether Qwen 3.8 27B Can Identify Games from Model Weights
Qwen 3.8 27B on my laptop can interpret model weights, too
- Who
- Simon Willison
- What
- tests whether Qwen 3.8 27B can identify games from raw neural network weights without tools
- When
- Quote
I’ve been playing with one of the examples from the paper and I’m delighted to share that Qwen 3.8 27B running on my laptop can do this as well!
What happened
Simon Willison reports that Qwen 3.8 27B running on his laptop identified Pac-Man, Pong, and Breakout from neural network weights, but mistook Freeway for a Pac-Man-like game. The prompts came from a paper’s accompanying GitHub repository and asked models to identify which grid game a policy network had learned, with no tools or code execution. Willison distinguishes this task from using code to update models and says the results do not justify claims that source code is becoming obsolete.
Read next
See the full storySimon Willison Tests Qwen 3.8 27B on Game Identification from Model WeightsThe story so far
Simon Willison found that Qwen 3.8 27B, running on his laptop, identified Pac-Man, Pong, and Breakout from neural network weights, but mistook Freeway for a Pac-Man-like game. The prompts asked which grid game a policy network had learned, using examples from a paper’s GitHub repository; the model had no tools or code execution. Willison said the test does not show that source code is becoming obsolete. He distinguished identifying a game from learned weights from using code to update models.
Keep up with AI, one email a day
Model releases, price changes and compute developments, with original sources. Free, confirmed by email, and easy to unsubscribe. Read a past issue · 2026-11-03