Simon Willison实测本地Qwen 3.8 27B可解读神经网络权重
Start here
2026年11月4日，技术开发者Simon Willison在本地笔记本上运行Qwen 3.8 27B模型，复现了权重空间编程相关论文的实验。测试结果证实，该模型在完全不使用任何代码或外部工具的情况下，仅凭输入的纯浮点权重文本，就成功推断出了对应策略网络所训练的具体游戏。该测试验证了该模型具备仅依靠权重原始数据便能直接理解并分析神经网络特征的能力。
AI-generated from reports · Updated 1 day ago
Report timeline
Follow the reports to see different sides of the story.
- Simon WillisonSelectedSimon Willison Confirms Qwen 3.8 27B Can Locally Interpret Neural Network Weights
Running Qwen 3.8 27B locally on a laptop, Simon Willison replicated experiments from a paper on weight-space programming, confirming the model can infer the specific game a policy network was trained on using only raw floating-point weight text, without relying on any code or tools.
Story heat trend
Current heat 4·Comparable peak 10（Nov 4 01:00）·Comparable change, last 24 hours-51%
Trends compare only the same parties observed continuously, so the scope can be narrower than current heat counts. Move the pointer or tap the chart to see hourly heat; use the left and right arrow keys on a keyboard.