self-consistency
Zero-cost epistemic uncertainty quantification & hallucination detection for LLMs (90,000x faster than Semantic Entropy)
Best-of-N 1.1 — sample a frozen language model N times and select the best answer. 45.3% to 66.5% on GSM8K with a 0.5B model, no training. Eight models measured, all 25,600 trajectories published.
Black-box AI reliability certification via self-consistency sampling and conformal calibration
