paper-with-me

홈 › Papers

VERDI: Single-Call Confidence Estimation for Verification-Based LLM Judges via Decomposed Inference

2026-05-11 · Jasmine Qi, Danylo Dantsev, Muyang Sun arxiv

LLM-as-Judge systems are widely deployed for automated evaluation, yet practitioners lack reliable methods to know when a judge's verdict should be trusted. Token log-probabilities, the standard post-hoc confidence signal, are unavailable for many commercial LLMs and, even when accessible, saturate above 0.999 with structured JSON output. We introduce VERDI (VERification-Decomposed Inference), a method that extracts confidence from the reasoning trace a structured judge already produces, with no additional inference calls. VERDI decomposes each verification-style evaluation into sub-checks and derives three structural signals: Step-Verdict Alignment, Claim-Level Margin, and Evidence Grounding Score. We combine them with Platt-scaled logistic regression. On three public benchmarks, VERDI achieves AUROC 0.72-0.91 on GPT-4.1-mini and 0.66-0.80 on GPT-5.4-mini. On Qwen3.5-4B/9B/27B, where answer-token logprobs are anti-calibrated (higher confidence on errors, AUROC 0.32-0.49), VERDI achieves 0.56-0.70. We additionally validate on a production system with eight rubrics (AUROC 0.73-0.88 on factual rubrics), demonstrate cross-model transfer (AUROC 0.66-0.69), and show that a 33M-parameter NLI (Natural Language Inference) model provides a scalable alternative to regex extraction.

📄 PDF Abstract BibTeX arXiv:2605.11334

Code (0)

등록된 구현이 없습니다.

Tasks

Natural Language Inference

Similar Papers 제목 키워드 기반

SVR: Self-Verifying Refinement via Joint Verdict-Confidence Reinforcement Learning for Adaptive Test-Time Compute

2026-07-30 · Hongyu Chen, Liang Lin, Guangrun Wang arxiv

Scaling test-time computation can improve language-model reasoning, but uniform budgets waste computation on easy inputs, while verifier-guided refinement relies on external feedback. We introduce Self-Verifying Refineme…

Mathematical ReasoningReinforcement Learning

Can Legal AI Know When It Is Wrong? And Do Students Know When It Is?

2026-08-21 · Angel Mary John, Vipin Kumar Singh, Jerrin Thomas Panachakel arxiv

Integrating Large Language Models (LLMs) into the Indian judiciary promises access to justice but introduces severe risks. We identify the 'inertia of confidence'--an overconfidence phenomenon analogous to the Dunning-Kr…

Cherry-pick Override: Unsafe Directional Commitment in LLM Judges under Mixed Evidence

2026-06-05 · Haoran Xu arxiv

LLM judges increasingly turn verdicts into system commitments. Under mixed evidence (claims with both supporting and refuting sources) this is unsafe: when the schema exposes CONFLICTING as the authorized non-directional…

From Verdict to Process: Agentic Reinforcement Learning for Multi-Stage Fact Verification

2026-06-11 · Rongxin Yang, Shenghong He, Siyuan Zhu, Chao Yu arxiv

Recent approaches combining Large Language Models (LLMs) with retrieval-augmented reasoning have shown promise for automated fact verification. To process complex claims, these verification pipelines typically execute mu…

Reinforcement LearningFact VerificationAnswer Generation

Evergreen: Efficient Claim Verification for Semantic Aggregates

2026-04-28 · Alexander W. Lee, Benjamin Han, Shayak Sen, Sam Yeom 외 arxiv

With recent semantic query processing engines, semantic aggregation has become a primitive operator, enabling the reduction of a relation into a natural language aggregate using an LLM. However, the resulting semantic ag…