AITOPAI热报
Event ·Multi-source

GPT-6 Astra 直接改底层参数修复 Qwen3.5-0.8B 死循环

Qwen3.5-0.8B 存在死循环缺陷。GPT-6 Astra 在不重新训练的情况下，仅修改该模型 0.016% 的底层参数完成修复。修复后数学测试成绩从 7.9% 提升到 48.1%，指令遵循能力提升约三倍。这展示了用前沿模型直接读写神经网络参数来调试小模型的路径。

Current read

GPT-6 Astra 只改了 Qwen3.5-0.8B 万分之十六的参数，没重训就把数学成绩从 7.9% 拉到 48.1%，挺离谱的实验。

11 sources58° AI heatLast updated 2026/11/3 01:43:17

Evidence

11 sources
Related source 3Thomas Wolf
View source
Related source 4rohanpaul_ai
View source
Related source 5The Rundown AI
View source
Related source 8shao__meng
View 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