AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

Building Reliable Long-Form Generation via Hallucination Rejection Sampling

Lin Li, Georgia Channing, Suhaas M Bhat, Gabriel Davis Jones, Yarin Gal

ICML 2026 · Large language models › Language, safety and interpretability · approach: Statistical frameworks for LLM evaluation · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
apply SHARS (Ours-Prec) to greedy decodingQwen3-32B · FactScorefactual precision (%): 52.4 -> 78.4Improved
apply SHARS (Ours-Prec) to greedy decodingLlama3.1-8B · FactScorefactual precision (%): 53.7 -> 78.4Improved
apply SHARS (Ours-Prec) to greedy decodingQwen3-4B · FactScorefactual precision (%): 50.0 -> 74.0Improved
apply SHARS (Ours-Prec) to greedy decodingLlama3.1-8B · FactScoresupported claims: 6.7 -> 5.0Hurt
apply SHARS (Ours-Prec) to greedy decodingQwen3-32B · FactScore (200-word constraint)factual precision (%): 58.0 -> 77.9Improved
use semantic entropy instead of token-level naive entropy (ablation)SHARS · FactScorefactual precision (%): 70.1 -> 78.4Improved
add SHARS (Ours-Prec) to FactAlignFactAlign Llama3-8B-Instruct · FactScore (no length constraint)factual precision (%): 53.1 -> 80.6Improved
add SHARS (Ours-Prec) to FactAlignFactAlign Llama3-8B-Instruct · FactScore (200-word constraint)factual precision (%): 55.4 -> 79.1Improved
add SHARS to greedy decodingQwen3-32B · LongFact (no length constraint)factual precision (%): 93.0 -> 94.6Improved
add rewriting to Following sampling (ablation)SHARS · FactScore ablationresponse rate (%): 54.4 -> 91.8Improved
add rewriting to Following sampling (ablation)SHARS · FactScore ablationfactual precision (%): 73.5 -> 69.4Hurt
use Following sampling instead of Temperature sampling (with rewriting) (ablation)SHARS · FactScore ablationfactual precision (%): 64.8 -> 69.4Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy