In-Context Meta Learning Induces Multi-Phase Circuit Emergence

Gouki Minegishi, Hiroki Furuta, Shohei Taniguchi, Yusuke Iwasawa, Yutaka Matsuo

Track: Long Paper Track (up to 9 pages)

Keywords: Mechanistic Interpretability, In-Context Learning, Circuits

TL;DR: We uncover how Transformers gain meta-learning capabilities by analyzing their internal circuits in an extended copy task, revealing multiple distinct phase transitions that explain in-context learning beyond simple copying.

Abstract: Transformer-based language models exhibit In-Context Learning (ICL), where predictions are made adaptively based on context. While prior work links induction heads to ICL through phase transitions, this can only account for ICL when the answer is included within the context. However, an important property of practical ICL in large language models is the ability to meta-learn how to solve tasks from context, rather than just copying answers from context; how such an ability is obtained during training is largely unexplored. In this paper, we experimentally clarify how such meta-learning ability is acquired by analyzing the dynamics of the model’s circuit during training by extending the copy task from previous research to an In-Context Meta Learning setting, where models must infer tasks from examples to answer queries. Interestingly, in this setting, we find that there are multiple phases in the process of acquiring such abilities, and that a unique circuit emerges in each phase, contrasting with the single-phase transition in induction heads. The emergence of such circuits can be related to several phenomena known in large language models, and our analysis lead to a deeper understanding of the source of the Transformer’s ICL ability.

Submission Number: 38