AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

Beyond Next Token Probabilities: Learnable, Fast Detection of Hallucinations and Data Contamination on LLM Output Distributions

Guy Bar-Shalom, Fabrizio Frasca, Derek Lim, Yoav Gelberg, Yftah Ziser, Ran El-Yaniv, Gal Chechik, Haggai Maron

AAAI 2026 · Large language models › Language, safety and interpretability · approach: Data and hallucination detection · 7 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add the full token distribution sequence (LOS-Net) instead of only actual token probabilities and rank (ATP+R-Transf.) (ablation)LOS-Net · IMDB hallucination detection, Mistral-7b-instructtest AUC: 89.64 -> 94.73Improved
add the full token distribution sequence instead of ATP and rank only (ablation)LOS-Net · BookMIA data contamination detection, Pythia-6.9bAUC: 79.59 -> 90.71Improved
add the full token distribution sequence instead of ATP and rank only (ablation)LOS-Net · HotpotQA hallucination detection, Llama3-8b-instructtest AUC: 66.72 -> 72.60Improved
increase top-K retained probabilities from 10 to 1000 (ablation)LOS-Net · HotpotQA, Mistral-7b-instructtest AUC: 71.82 -> 72.92Improved
use only K=10 top probabilities (API-limited access) (ablation)LOS-Net · HotpotQA, Mistral-7b-instructtest AUC: 72.92 -> 71.82No clear effect
learned LOS-Net instead of heuristic gray-box scores and prompting-based methodsLOS-Net · IMDB hallucination detection, Mistral-7b-instructtest AUC: around 31 AUC units over best baselineImproved
learned LOS-Net instead of reference-free heuristics (Min-K%, Min-K%++, Loss)LOS-Net · BookMIA, Pythia-6.9bAUC: 68.78 -> 90.71Improved
zero-shot transfer of LOS-Net to unseen LLMs without fine-tuningLOS-Net · BookMIA cross-LLMAUC: best among reference-free in 10/12 casesImproved
zero-shot cross-LLM application for hallucination detectionLOS-Net · hallucination detectionAUCHurt
fine-tune a pretrained LOS-Net on other datasets/LLMs instead of training from scratchLOS-Net · hallucination detection transfertest AUC: better in 16/18 casesImproved
train on Llama3-8b and test on Mistral-7b with fine-tuningLOS-Net · IMDB hallucination detection, cross-LLMtest AUC: around 27 AUC units over best probability-based baselineImproved
learned LOS-Net vs white-box Activation ProbeLOS-Net · hallucination detection, 6 dataset/LLM combostest AUC: better in 2/6, similar in 3/6Mixed

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy