Reproduce “Measuring Mechanistic Uncertainty in Large Language Models via Causal Robustness” 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
Aneesh Komanduri, Majid Afshar, Timothy A. Miller, Yanjun Gao
The authors introduce mechanistic uncertainty by measuring the minimum internal steering magnitude, α*, that changes a model’s answer semantically. They use Jacobian Lens to read consensus concepts from activations and intervene across selected layers. On TriviaQA, Qwen2.5-7B’s margin scores incorrect answers with 75.14% AUROC. The measure reflects robustness to this steering procedure, not universal reliability or truth.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.