Get Started
Home
Topics
Search
Library
Research questionHow can black-box LLM APIs detect hallucinations without trusted context while limiting false alarms?Without a reference document, an LLM may produce consistently confident errors or varied answers whose meaning is difficult to assess. Detection must therefore distinguish these failure patterns while avoiding excessive referrals for human review.
AI
Alignment & Safety
Evaluation & Benchmarks
Machine Learning
Natural Language Processing
Latest papersRecent research connected to this question, newest first.From Tokens to Semantics: Leveraging Complementary Signals for Hallucination Detection in Black-Box LLMsThe source studies semantic entropy, token log-probability uncertainty, and combinations of these signals using black-box APIs. Evidence comes from seven text and multimodal benchmarks across four language models, with performance examined under false-positive-rate budgets from 1% to 15% and varying generation and calibration choices.research paper · Sep 7, 2026
Related questions
How can we reliably detect when an LLM response is unsupported by its reference documents?How can sampling-only diagnosis identify whether an individual LLM hallucination comes from retrieval, training, or autoregressive commitment?How can hallucination detectors verify claims and localize unsupported spans without costly separate pipelines?How can black-box LLMs resist jailbreaks without weight access or retraining while preserving benign-query utility?