Detecting Errors in Large Language Model Responses: The Incremental Predictive Value of Hidden States
Although large language models support a wide range of tasks, hallucinations remain a major obstacle to user trust.
ProofPaper ↗
Key points
- When a model generates an answer, both its internal processing and final output vary with the question.
- Internal states are often expected to offer richer signals for hallucination detection.
- We integrate output-side information available at inference time into a combined baseline and compare detection performance before and after adding hidden states under identical data splits and evaluation rules.
- Experiments cover open-ended generation and constrained multiple-choice tasks across different models.
Sources (1)
- [1]Detecting Errors in Large Language Model Responses: The Incremental Predictive Value of Hidden StatesarXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 20, 01:12 AM
Although large language models support a wide range of tasks, hallucinations remain a major obstacle to user trust.
When a model generates an answer, both its internal processing and final output vary with the question.
Extractive summary: sentences quoted from the sources.
Before this
- Oct 20, 2026Matrix-Game 4.0: Harnessing Video Models for Real-Time Playable Game Worlds
- Oct 19, 2026Introducing Clairic, a new System One class of extraction models
- Oct 19, 2026Push-based KV transfer to improve TTFT and TPOT for disaggregated inference in vLLM
- Oct 19, 2026NVIDIA Cosmos3-H-Dreams: Bringing Clinical Surgery to Life Through Generative Simulation
- Oct 19, 2026From Black Box to Controller: Inference-Time Steering of LLM via Feature Amplification
- Oct 18, 2026TIRE: Top-Down Formula Instantiation with Reliable Execution for Physics Problem Synthesis