aiqbench analysis
This research addresses the critical challenge of detecting hallucinations in black-box LLMs by fusing semantic entropy with token-level uncertainty, offering a robust solution when reference documents are unavailable. The study demonstrates that combining these complementary signals significantly improves detection accuracy across diverse benchmarks without requiring universal superiority from any single method. For practitioners deploying high-stakes AI systems, this work provides actionable strategies to balance false positives and missed fabrications in real-world workflows.
From Tokens to Semantics: Leveraging Complementary Signals for Hallucination Detection in Black-Box LLMs
arXiv:2609.02679v1 Announce Type: new Abstract: When LLMs support public-facing or high-stakes workflows, missed fabrications can harm users and institutions, while false alarms consume limited human-review capacity. When no trusted context or reference document is available, we study two signals accessible through black-box model APIs: semantic entropy, which measures disagreement among sampled response meanings, and uncertainty derived from token log-probabilities. Their failure modes can be complementary: semantic entropy becomes uninformative when responses form one semantic cluster, while token uncertainty can miss consistently confident errors. We extend token-based uncertainty detection by aggregating token-level signals across sampled responses through our TopK method, evaluate the hybrid CoCoA method, which combines target-response uncertainty with semantic dissimilarity, and propose and study two supervised methods: Gated, which routes single-cluster cases to an aggregated-token-feature classifier, and Stacked, which learns jointly from semantic uncertainty and broader token features. We evaluate seven benchmarks, including five public benchmarks (four text datasets and multimodal handwritten-cheque extraction) and two constructed benchmarks (Financial Summaries and Long-Text QA), using four language models. In our evaluation across models and datasets, Stacked gave the best performance in nearly half of the cases, while TopK and CoCoA remain competitive without supervised training labels, although their thresholds require careful calibration. No method is universally strongest. We therefore evaluate performance at false-positive-rate budgets from 1% to 15%, assess their sensitivity to generation and calibration choices, and examine variation across dataset characteristics.
Abstract:When LLMs support public-facing or high-stakes workflows, missed fabrications can harm users and institutions, while false alarms consume limited human-review capacity. When no trusted context or reference document is available, we study two signals accessible through black-box model APIs: semantic entropy, which measures disagreement among sampled response meanings, and uncertainty derived from token log-probabilities. Their failure modes can be complementary: semantic entropy becomes uninformative when responses form one semantic cluster, while token uncertainty can miss consistently confident errors. We extend token-based uncertainty detection by aggregating token-level signals across sampled responses through our TopK method, evaluate the hybrid CoCoA method, which combines target-response uncertainty with semantic dissimilarity, and propose and study two supervised methods: Gated, which routes single-cluster cases to an aggregated-token-feature classifier, and Stacked, which learns jointly from semantic uncertainty and broader token features. We evaluate seven benchmarks, including five public benchmarks (four text datasets and multimodal handwritten-cheque extraction) and two constructed benchmarks (Financial Summaries and Long-Text QA), using four language models. In our evaluation across models and datasets, Stacked gave the best performance in nearly half of the cases, while TopK and CoCoA remain competitive without supervised training labels, although their thresholds require careful calibration. No method is universally strongest. We therefore evaluate performance at false-positive-rate budgets from 1% to 15%, assess their sensitivity to generation and calibration choices, and examine variation across dataset characteristics.
Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
arXiv:2609.02679 [cs.CL]
(or arXiv:2609.02679v1 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2609.02679 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 324 words
Related coverage
Deterministic Hallucination Detection in Medical VQA via Confidence-Evidence Bayesian Gain
arXiv · Jul 21, 2026
Building trust in AI agents: A production-ready evaluation framework with Amazon Bedrock AgentCore
AWS · Oct 16, 2026
Layer-Resolved Optimal Transport for Hallucination Detection in NMT and Abstractive Summarization
arXiv · Jun 12, 2026
Stop Token-Incinerating Your Risk Engine
Neo4j · Sep 15, 2026