LLMs Can Annotate Attribution Graphs

大语言模型能够标注归因图

Circuit tracing is an exciting technique for revealing the internal computation of language models, but it requires a time-intensive manual step of grouping individual features or MLP neurons into supernodes. 电路追踪（Circuit tracing）是一种揭示语言模型内部计算过程的令人兴奋的技术，但它需要一个耗时的人工步骤，即将单个特征或 MLP 神经元分组为“超节点”（supernodes）。

We present a simple pipeline for automating this step: directly presenting feature descriptions to a language model that groups them into supernodes. 我们提出了一种自动化该步骤的简单流程：直接将特征描述提供给大语言模型，由其将这些特征归类为超节点。

Using automated interpretability metrics, we confirm that supernodes generated by our pipeline are as interpretable as those generated by human annotators. 通过使用自动化可解释性指标，我们证实了由我们的流程生成的超节点与人工标注者生成的超节点具有相同的可解释性。

On a two-hop Capitals task, our pipeline recovers a supernode corresponding to the intermediate hop in 97 of 100 prompts. 在“两跳首都”（two-hop Capitals）任务中，我们的流程在 100 个提示词中的 97 个里，成功恢复了对应于中间跳跃步骤的超节点。

Finally, we present a simple proof of concept using our pipeline for open-ended exploration, where we automatically annotate 1000 attribution graphs from Wikipedia prompt completions and then use an LLM judge to flag interesting graphs worth human review. 最后，我们展示了一个使用该流程进行开放式探索的简单概念验证：我们自动标注了来自维基百科提示补全的 1000 个归因图，然后使用一个大语言模型裁判来标记出值得人工审查的有趣图表。

We hope this work demonstrates that even simple automation can produce meaningful attribution graph annotations, motivating further work on automated circuit tracing. 我们希望这项工作能够证明，即使是简单的自动化也能产生有意义的归因图标注，从而推动关于自动化电路追踪的进一步研究。