Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper formulates hallucination onset detection as a quickest change detection problem, leveraging a first-order Markov model to establish a theoretical lower bound on detection delay. The authors demonstrate that a causal recurrent labeler, functioning as a CUSUM with a learned increment, significantly outperforms a linear per-token baseline, detecting hallucinations in 11-13 tokens compared to 31. The study reveals that the advantage primarily stems from improved per-token scoring rather than temporal accumulation, highlighting the limitations of traditional classification metrics in capturing detection delays.
A causal recurrent labeler can detect hallucinations nearly three times faster than traditional methods, revealing critical insights into detection delay dynamics.
Token-level hallucination detectors are evaluated as classifiers, by AUC over all tokens, yet a streaming monitor is judged by its reaction time: the number of tokens that pass between the onset of a hallucination and the alarm. We formulate hallucination onset detection as a quickest change detection problem. A first-order Markov model of the latent faithful/hallucinated state, validated on RAGTruth, places the task inside classical change-point theory and yields Lorden's lower bound on detection delay: about 1.3 tokens at a false-alarm rate of 0.01. We then show that a causal recurrent labeler acts as a CUSUM with a learned increment; at a matched false-alarm rate it detects in 11-13 tokens, against 31 for a linear per-token baseline, and a controlled decomposition attributes most of this advantage to a better per-token score rather than to temporal accumulation. An information-rate optimality theorem of Donsker-Varadhan type explains the remaining order-of-magnitude gap: the learned score realizes only 1/4.5 of the divergence the features carry, a deficit that recalibration cannot remove, with the remainder a finite-horizon effect. Classification metrics conceal this delay structure; sequential analysis makes it measurable