Semantic Entropy as a Scalable Hallucination Detection Signal in Large Language Models: Theoretical Grounding, Calibration, and Downstream Task Benchmark
2026· Princeton Journal of Pre-Collegiate Research· 0 citations
TL;DR
These findings contribute empirically to machine learning and natural language processing and carry actionable implications for the design of programs and policies targeting LLM-powered QA system reliability gating, medical/legal hallucination risk reduction, and uncertainty-aware RAG system design.
Abstract
This study investigates semantic entropy as a computationally tractable hallucination detection signal in large language models, grounding the method in information-theoretic terms and benchmarking calibration and F1 on five factual QA datasets within the context of machine learning and natural language processing, an area of growing scientific importance given its implications for LLM-powered QA system reliability gating, medical/legal hallucination risk reduction, and uncertainty-aware RAG system design. Using semantic entropy computed as H_sem = -sum p(c)*log p(c) over meaning-equivalent clusters of 20 sampled responses, with calibration (ECE) and hallucination detection F1 compared to logit-based, embedding-variance, and self-consistency baselines, we examine sampling-induced variation in semantically equivalent responses capturing epistemic uncertainty about factual claims, with high semantic entropy indicating model uncertainty signaling potential hallucination independent of confidence calibration artifacts in 24,200 question-answer pairs across 5 benchmarks (TriviaQA 5k, NQ 5k, SQuAD 5k, BioASQ 4.2k, HalluBench 5k) evaluated across 3 LLMs with 20 sampled responses per question = 1.45 million total response tokens drawn from inference via API (GPT-4o, Claude 3.5 Sonnet) and local A100 cluster (Llama 3.1-70B) with semantic clustering via SentenceBERT cosine similarity threshold 0.84 for equivalence. Results indicate that semantic entropy achieves mean AUC 0.82 across 5 benchmarks and 3 LLMs (vs. 0.68 self-consistency and 0.64 logit-based), with ECE 0.06 vs. 0.18 for logit confidence; BioASQ domain shows lowest AUC (0.74) due to synonym-rich medical terminology clustering failures (p < 0.001), with mean AUC 0.82 vs. 0.68 self-consistency; ECE 0.06 vs. 0.18 as the primary quantitative benchmark. Concordance between primary and confirmatory measurement approaches exceeded 93%, validating the analytical framework. These findings contribute empirically to machine learning and natural language processing and carry actionable implications for the design of programs and policies targeting LLM-powered QA system reliability gating, medical/legal hallucination risk reduction, and uncertainty-aware RAG system design.
In this paper, we present an empirical analysis of hallucination behavior in small language models (1.5B–7B parameters), addressing the question of whether such failures are stochastic or structured. Our analysis of 4,500 responses from three instruction-tuned models, evaluated on TruthfulQA and HaluEval, reveals that...
Kushagra Agrawal, Yu-Ming Feng, Hang-Jun Che et al.· International Conference on...· 0 citations
These findings show that aggregate detection metrics conceal persistent, model dependent heterogeneity in language model failures and motivate regime conditioned evaluation.
P. Darshan, A. Pranav, T. SravanKarthick et al.· arXiv.org· 0 citations
Bounded margins mitigate confident hallucinations during post-training, implemented through an entropy-dependent margin bound in direct preference optimization (DPO) and shown to mitigate confident hallucinations during post-training.
Hallucinations are false or unverifiable outputs generated by Large Language Models (LLMs) that reduce their trustworthiness in critical applications such as healthcare, education, and research. This paper proposes a hybrid framework that unifies three complementary verification signals for the detection and reduction...
Radhika K. R., Sumit Gupta, Nagesh H. Shenoy et al.· 2026 Forth International Con...· 0 citations
A thorough comparative analysis using a variety of metrics of ten advanced LLMs shows a fundamental trade-off between response completeness and factual reliability, intensified by prompt complexity.
Osama Shabih, Dayam Nadeem, S. Sohail et al.· 0 citations
Large vision-language models (LVLMs) are increasingly used for multimodal search, open-ended question answering, document understanding, and generated content analysis, yet they still produce claims that are fluent but unsupported by visual evidence. Existing hallucination mitigation methods often rely on external mult...
Tong Zhang, Qing-Qing Ren, Hua-Ling Liu et al.· Proceedings of the 35th ACM...· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.