Event ·Multi-source

Simon Willison 实测 GPT-6 Astra 直接解读模型权重的方法

一篇新论文声称 GPT-6 Astra 能直接解读模型权重，即通过充满浮点数的提示词读取模型内部信息。Simon Willison 在自己的笔记本上用 Qwen 3.8 27B 复现了该实验。经过 45 分钟以上的运行，Qwen 3.8 27B 也实现了同样效果。

Current read

有人用 Qwen 3.8 27B 在笔记本上复现了 GPT-6 Astra 读权重的方法，跑了 45 分钟，过程和结论都写在博客里。

11 sources33° AI heatLast updated 2026/11/3 16:31:05

Evidence

11 sources

Primary sourceSimon Willison

2026/11/3 16:31:05View source

Related source 1kimmonismus

2026/11/2 19:48:50View source

Related source 2IT之家

2026/11/3 02:00:55View source

Related source 3歸藏(guizang.ai)

2026/11/3 03:30:00View source

Related source 4rohanpaul_ai

2026/11/2 12:45:11View source

Related source 5Jerry Liu

2026/11/2 19:44:32View source

Related source 6Thomas Wolf

2026/11/2 20:15:21View source

Related source 7The Rundown AI

2026/11/2 20:17:39View source

Related source 8elvis

2026/11/2 20:31:05View source

Related source 9orange.ai

2026/11/2 23:30:19View source

Related source 10宝玉

2026/11/3 00:06:53View source

Conflict check

Dedup data only shows that these sources discuss the same event, not that they agree. There is no structured supporting or refuting evidence yet, so no conflict is inferred automatically.

Open the story