Reproduce “A Circuit for In-Context Learning” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Andrew Gao, William Gao
The authors explain in-context learning for two-class sentence classification as gated associative lookup: writers store sentence class and label identity at demonstration label positions, and readers retrieve matching labels when a repeat signal opens the read. Brake heads counter recency. In Qwen3-4B, reader-output patching carried 60% of a class-flip margin change; evidence is limited to three models and this task.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.