Qwen 3.8 27B on my laptop can interpret model weights, too
Yesterday's hot paper was Sparks of Weight-Space Programming: LLMs can natively read & write neural network weights by Rikiya Takehi, Jayden Teoh, Alex Zhang, and Omar Khattab. The paper found that frontier models, in particular GPT-6 Astra, can both interpret and modify the raw weights of a machine learning model - literally a text description of the model containing a bunch of floating point…
The recent paper Sparks of Weight-Space Programming revealed that large language models, such as GPT-6 Astra, can interpret and modify the raw weights of a machine learning model. Inspired by this, the author tested Qwen 3.8 27B on their laptop and found it capable of interpreting model weights as well. The paper presents the first question as: Can LLMs understand model behaviors from weights?
To test this, weights from four neural networks trained to play simplified versions of classic computer games were fed to an LLM, which had to identify the original game without any tools or code execution. The author successfully ran this test using Kimi K3.1 Pro via OpenRouter and Qwen 3.8 27B on their laptop. Qwen accurately identified Breakout, Pong, and Pac-Man, but struggled with the Freeway game.
Written by urgent.news from Hacker News's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
This story
This is one outlet's version. Read the fullest account.
- Qwen 3.8 27B on my laptop can interpret model weights, too simonwillison.net
- Qwen 3.8 27B on my laptop can interpret model weights, too substack.com
