Reproduce “The Trilemma of Truth in Large Language Models” 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

Germans Savcisens, Tina Eliassi-Rad

The authors introduce sAwMIL, a probe that classifies statements as true, false, or neither by learning from token-level representations across a statement, then uses conformal prediction sets to express uncertainty. Across 16 models and three English domains, it transfers better than tested binary and last-token probes. Steering yields a 0.439 NIE, but “neither” relies on synthetic entities, not verified unknowns.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.