精选6504:17Rohan Paul模型发布/更新多源精选 ×6

MIT研究称GPT-6 Astra可微调Qwen3.5权重提升数学能力

MASSIVE. MIT researchers showed that GPT-6 Astra can improve another AI model by…

MASSIVE. MIT researchers showed that GPT-6 Astra can improve another AI model by changing its weights directly, without training.

大规模。麻省理工学院研究人员表明，GPT-6 Astra 可以通过直接更改权重来提升另一个 AI 模型的性能，而无需进行训练。

Here, GPT-6 Astra raised Qwen3.5-0.8B from 7.9% to 48.1% on MATH-500 by editing 0.016% of its weights, without training.

在此案例中，GPT-6 Astra 通过编辑 Qwen3.5-0.8B 的 0.016% 权重（无需训练），将其在 MATH-500 数据集上的得分从 7.9% 提升至 48.1%。

Astra got Qwen’s weights in a Python notebook, plus a budget of 50 test runs on practice math problems. It saw the model kept repeating itself until it hit its length limit, traced that habit to a single internal signal written by 2 of its layers, and subtracted that signal out of the weights.

Astra 在一个 Python 笔记本中获取了 Qwen 的权重，并拥有 50 次练习数学问题的测试运行预算。它观察到模型在达到长度限制前会不断重复自身，并将这一习惯追溯至其两个层写入的一个内部信号，随后将该信号从权重中减去。

With the loop gone, the model could finish answers it was already capable of solving.

随着循环被消除，模型能够完成那些它原本就有能力解答的答案。

That single targeted edit changed 0.016% of the weights and lifted math accuracy from 7.9% to 48.1%.

这次单一的针对性编辑改变了 0.016% 的权重，使数学准确率从 7.9% 提升至 48.1%。