MAEM 方法用 RL 生成文本样例，解读 Qwen3.6-27B 残差流方向
Towards interpreting any residual stream direction by scaling activation inversion
一群研究者做了个叫 MAEM 的元模型，能把 Qwen3.6-27B 里任意残差流方向翻译成人能读懂的文本，还比语料搜索和 NLA 更准，解释能用一次前向传播验证，可解释性方向值得看。
论文提出 Maximally-Activating Example Metamodels（MAEM），用一个语言模型把目标模型的残差流方向映射为能强激活该方向的文本样例，训练目标是最大化方向与生成文本激活的余弦相似度，全程无需标注。在 Qwen3.6-27B 上的实验显示，MAEM 反演 held-out 激活的效果超过语料搜索（1000 万 token）和自然语言自编码器（NLA），在多数 held-out 语言和领域上胜过语料搜索，并激活 89% 的 held-out SAE 特征。它还能像 NLA 一样识别 steering vector 背后的概念，比语料搜索更常恢复 LoRA 后门的载荷，甚至能口头复述部分模型工作区的内部独白。与 NLA 的解释不同，MAEM 的样例可以在目标模型上用一次前向传播直接验证。
Towards interpreting any residual stream direction by scaling activation inversion
To interpret a direction in a language model's residual stream, we can look for text that strongly activates it. But when such text is found by scanning a corpus, it is limited to what that corpus contains. We develop Maximally-Activating Example Metamodels (MAEMs), language models which map residual stream directions from a subject model to text exemplars that strongly activate those directions. Our objective is to maximise the cosine similarity of a supplied residual stream direction with the activations of the text generated by the MAEM. This label-free RL objective helps to scale MAEMs on diverse sources of residual stream directions. Empirically, we find the resultant metamodel produces informative exemplars: a Qwen3.6-27B MAEM inverts held-out activations better than corpus search or a natural-language autoencoder (NLA), beats corpus search (at 10M tokens) on a majority of held-out languages/domains, and activates 89% of held-out SAE features. It also identifies concepts behind steering vectors about as well as NLAs, recovers the payloads of LoRA backdoors more often than corpus search, and can even verbalise some model workspace inner monologue. Unlike explanations produced by other metamodels (e.g. NLAs), MAEM exemplars can be verified with a forward pass on the subject model, making the MAEM objective a grounded alternative for scaling the training of metamodels.