Primary sourceorange.ai
View sourceEvent ·Multi-source
GPT-6 Astra 直接改写 Qwen3.5-0.8B 权重,数学能力提升 6 倍
研究者 Rikiya Takehi 让 GPT-6 Astra 直接编辑 Qwen3.5-0.8B 的权重来提升数学能力。整个过程不使用梯度、不做训练,模型只改动了 0.016% 的权重,就修复了 0.8B 小模型陷入推理死循环的问题。改动后 Qwen3.5-0.8B 的数学能力提升最高达 6 倍,指令遵循能力也提升 3 倍。研究者将这种方法称为 Weight-Space Programming(权重空间编程),即前沿 LLM 原生读写神经网络权重。
Current read
有人让 GPT-6 直接手改 Qwen 小模型的权重,不改 99.98% 的参数,数学就涨了 6 倍,全程零训练零梯度,思路很邪门,推荐看看原文。
11 sources55° AI heatLast updated 2026/11/2 23:30:19
Evidence
11 sourcesRelated source 1Thomas Wolf
View sourceRelated source 2李继刚
View sourceRelated source 3Jerry Liu
View sourceRelated source 4The Rundown AI
View sourceRelated source 5elvis
View sourceRelated source 6shao__meng
View sourceRelated source 7歸藏(guizang.ai)
View sourceRelated source 8IT之家
View sourceRelated source 9rohanpaul_ai
View sourceRelated source 10Latent.Space
View sourceConflict 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.