A DETR-style probe distills multi-sample claim uncertainty into single-pass span detection and continuous Mixture-of-Beta scores, outperforming baselines on a new 293K-span benchmark.
hub
Semantic Entropy Probes: Robust and Cheap Hallucination Detection in LLMs
44 Pith papers cite this work, alongside 3 external citations. Polarity classification is still indexing.
44
Pith papers citing it
3
external citations · Pith
abstract
We propose semantic entropy probes (SEPs), a cheap and reliable method for uncertainty quantification in Large Language Models (LLMs). Hallucinations, which are plausible-sounding but factually incorrect and arbitrary model generations, present a major challenge to the practical adoption of LLMs. Recent work by Farquhar et al. (2024) proposes semantic entropy (SE), which can detect hallucinations by estimating uncertainty in the space semantic meaning for a set of model generations. However, the 5-to-10-fold increase in computation cost associated with SE computation hinders practical adoption. To address this, we propose SEPs, which directly approximate SE from the hidden states of a single generation. SEPs are simple to train and do not require sampling multiple model generations at test time, reducing the overhead of semantic uncertainty quantification to almost zero. We show that SEPs retain high performance for hallucination detection and generalize better to out-of-distribution data than previous probing methods that directly predict model accuracy. Our results across models and tasks suggest that model hidden states capture SE, and our ablation studies give further insights into the token positions and model layers for which this is the case.
hub tools
citation-role summary
baseline 2
citation-polarity summary
roles
baseline 2polarities
baseline 2representative citing papers
A between-subjects experiment (N=192) finds that token-level uncertainty increases agreement with LLM answers while relation-level uncertainty reduces external verification in medical decision tasks.
Verification prompting in LVLMs is input-dependent and risk-bearing; RSP selectively triggers it via pre-generation uncertainty to avoid performance degradation on easy cases.
Inducing artificial uncertainty on trivial tasks allows training probes that achieve higher calibration on hard data than standard approaches while retaining performance on easy data.
Conflict and hallucination in transformers are basin competition versus basin absence in hidden-state space; geometric margin detects them with zero false refusals while entropy cannot, and confident hallucinations scale as exp(-c/Δ̄).
Answer correctness and question answerability are separate axes: ordinary confidence tracks the first while hidden probes track the second, and a factorized dual-threshold policy certifies both risk budgets at higher correct-answer coverage.
Unsupervised MLP activation dispersion separates known from fabricated entities at AUROC 0.95–1.00 across Bielik scales, while factual reliability scales separately and refusals stay near zero.
Hidden-state probes can flag doomed LLM-agent episodes from the first round, and a recall-calibrated cascade of abort gates cuts generated tokens by up to 60% while preserving a chosen success-recall target.
A large-scale multilingual evaluation of LLM uncertainty estimation methods across 22 languages and 9 models finds that English reasoning closes the UE gap for low-resource languages and that optimal UE method choice depends on model scale.
Argus benchmark shows UQ method rankings for GUI grounding agents are stable within models across datasets but degrade across model classes and to closed-source vendors.
Grad Detect uses internal gradient patterns from one inference pass to predict LLM hallucinations and abstention, outperforming confidence and sampling baselines on Q&A benchmarks with most signal in the final five layers.
Agent systems lose uncertainty at decision handoffs, causing downstream over-trust; the paper proposes latent uncertainty as a carrier to preserve pre-commitment fragility across interfaces.
Density ridge scoring on 6D kinematic features from hidden states yields 5-20 point AUROC gains over Semantic Entropy and log-probability baselines for hallucination detection under a 200-query calibration protocol across six QA benchmarks and eight models.
Propagational Proxy Voting driven by letter entropy and centered reasoning embeddings beats majority by +2.24 pp on non-trivial MMLU-Pro questions without labels or training.
Knot estimates QA model sensitivity to candidate knowledge via subset counterfactual training and latent factor coverage, yielding unit rankings that outperform baselines without extra model calls.
Larger LLMs hallucinate more often despite having the correct concept available because instruction tuning causes probability mass to disperse across alternative surface forms instead of concentrating on one.
Trajectory geometry of MLP write-vectors, summarized by 11 scale-invariant features in a sparse linear probe, beats MSP for selective abstention in LLM multiple-choice QA.
REALISTA generates semantically coherent adversarial prompts via latent-space optimization over input-dependent editing directions, achieving stronger hallucination elicitation than prior realistic attacks on open-source and reasoning LLMs.
Temporal knowledge drift is encoded as a geometrically orthogonal direction in LLM residual streams, independent of correctness and uncertainty.
Probabilistic circuits detect LLM hallucinations as residual-stream anomalies with up to 99% AUROC and enable dynamic correction that raises truthfulness scores while cutting unnecessary output corruption.
Average token log-probability provides a zero-shot confidence signal for small LLMs that matches supervised baselines in-distribution and outperforms them out-of-distribution, with a new retrieval-conditional variant improving further at lower latency.
LLM token rank-frequency distributions converge to a shared Mandelbrot distribution across models and domains, enabling a microsecond-scale statistical primitive for provenance verification and black-box anomaly triage.
Temporal difference calibration aligns uncertainty estimates in vision-language-action models with their value functions for better sequential performance.
Unsupervised single-generation confidence calibration for reasoning LLMs via offline self-consistency proxy distillation outperforms baselines on math and QA tasks and improves selective prediction.
citing papers explorer
-
SpanUQ: Span-Level Uncertainty Quantification for Large Language Model Generation
A DETR-style probe distills multi-sample claim uncertainty into single-pass span detection and continuous Mixture-of-Beta scores, outperforming baselines on a new 293K-span benchmark.
-
Not All Uncertainty Is Equal: How Uncertainty Granularity Shapes Human Verification in LLM-Assisted Decision Making
A between-subjects experiment (N=192) finds that token-level uncertainty increases agreement with LLM answers while relation-level uncertainty reduces external verification in medical decision tasks.
-
Risk-aware Selective Prompting for Hallucination Mitigation in Large Vision-Language Models
Verification prompting in LVLMs is input-dependent and risk-bearing; RSP selectively triggers it via pre-generation uncertainty to avoid performance degradation on easy cases.
-
Inducing Artificial Uncertainty in Language Models
Inducing artificial uncertainty on trivial tasks allows training probes that achieve higher calibration on hard data than standard approaches while retaining performance on easy data.
-
Attractor Geometry of Transformer Memory: From Conflict Arbitration to Confident Hallucination
Conflict and hallucination in transformers are basin competition versus basin absence in hidden-state space; geometric margin detects them with zero false refusals while entropy cannot, and confident hallucinations scale as exp(-c/Δ̄).
-
Two Axes of LLM Abstention: Answer Correctness and Question Answerability
Answer correctness and question answerability are separate axes: ordinary confidence tracks the first while hidden probes track the second, and a factorized dual-threshold policy certifies both risk budgets at higher correct-answer coverage.
-
Does Bielik Know What It Doesn't Know? Activation Dispersion Separates Entity Familiarity from Factual Reliability Across Model Scale
Unsupervised MLP activation dispersion separates known from fabricated entities at AUROC 0.95–1.00 across Bielik scales, while factual reliability scales separately and refusals stay near zero.
-
Doomed from the Start: Early Abort of LLM Agent Episodes via a Recall-Controlled Probe Cascade
Hidden-state probes can flag doomed LLM-agent episodes from the first round, and a recall-calibrated cascade of abort gates cuts generated tokens by up to 60% while preserving a chosen success-recall target.
-
Estimating Uncertainty from Reasoning: A Large-Scale Study of Multi- and Crosslingual MCQA Performance in LLMs
A large-scale multilingual evaluation of LLM uncertainty estimation methods across 22 languages and 9 models finds that English reasoning closes the UE gap for low-resource languages and that optimal UE method choice depends on model scale.
-
Uncertainty Quantification for Computer-Use Agents: A Benchmark across Vision-Language Models and GUI Grounding Datasets
Argus benchmark shows UQ method rankings for GUI grounding agents are stable within models across datasets but degrade across model classes and to closed-source vendors.
-
Grad Detect: Gradient-Based Hallucination Detection in LLMs
Grad Detect uses internal gradient patterns from one inference pass to predict LLM hallucinations and abstention, outperforming confidence and sampling baselines on Q&A benchmarks with most signal in the final five layers.
-
Confidence Laundering in Agent Systems: Why Uncertainty Needs a Latent Carrier
Agent systems lose uncertainty at decision handoffs, causing downstream over-trust; the paper proposes latent uncertainty as a carrier to preserve pre-commitment fragility across interfaces.
-
Density Ridge Selective Prediction for LLM and VLM Hallucination Detection under Calibration Label Scarcity
Density ridge scoring on 6D kinematic features from hidden states yields 5-20 point AUROC gains over Semantic Entropy and log-probability baselines for hallucination detection under a 200-query calibration protocol across six QA benchmarks and eight models.
-
When Does Delegation Beat Majority? A Delegation-Based Aggregator for Multi-Sample LLM Inference
Propagational Proxy Voting driven by letter entropy and centered reasoning embeddings beats majority by +2.24 pp on non-trivial MMLU-Pro questions without labels or training.
-
Knowledge Dependency Estimation for Reliable Question Answering
Knot estimates QA model sensitivity to candidate knowledge via subset counterfactual training and latent factor coverage, yielding unit rankings that outperform baselines without extra model calls.
-
Hallucination as Commitment Failure: Larger LLMs Misfire Despite Knowing the Answer
Larger LLMs hallucinate more often despite having the correct concept available because instruction tuning causes probability mass to disperse across alternative surface forms instead of concentrating on one.
-
Reading Calibrated Uncertainty from Language Model Trajectories
Trajectory geometry of MLP write-vectors, summarized by 11 scale-invariant features in a sparse linear probe, beats MSP for selective abstention in LLM multiple-choice QA.
-
REALISTA: Realistic Latent Adversarial Attacks that Elicit LLM Hallucinations
REALISTA generates semantically coherent adversarial prompts via latent-space optimization over input-dependent editing directions, achieving stronger hallucination elicitation than prior realistic attacks on open-source and reasoning LLMs.
-
The Geometry of Forgetting: Temporal Knowledge Drift as an Independent Axis in LLM Representations
Temporal knowledge drift is encoded as a geometrically orthogonal direction in LLM residual streams, independent of correctness and uncertainty.
-
Hallucination as an Anomaly: Dynamic Intervention via Probabilistic Circuits
Probabilistic circuits detect LLM hallucinations as residual-stream anomalies with up to 99% AUROC and enable dynamic correction that raises truthfulness scores while cutting unnecessary output corruption.
-
Zero-Shot Confidence Estimation for Small LLMs: When Supervised Baselines Aren't Worth Training
Average token log-probability provides a zero-shot confidence signal for small LLMs that matches supervised baselines in-distribution and outperforms them out-of-distribution, with a new retrieval-conditional variant improving further at lower latency.
-
The Surprising Universality of LLM Outputs: A Real-Time Verification Primitive
LLM token rank-frequency distributions converge to a shared Mandelbrot distribution across models and domains, enabling a microsecond-scale statistical primitive for provenance verification and black-box anomaly triage.
-
Temporal Difference Calibration in Sequential Tasks: Application to Vision-Language-Action Models
Temporal difference calibration aligns uncertainty estimates in vision-language-action models with their value functions for better sequential performance.
-
Unsupervised Confidence Calibration for Reasoning LLMs from a Single Generation
Unsupervised single-generation confidence calibration for reasoning LLMs via offline self-consistency proxy distillation outperforms baselines on math and QA tasks and improves selective prediction.
-
Mind the Unseen Mass: Unmasking LLM Hallucinations via Soft-Hybrid Alphabet Estimation
SHADE adaptively combines coverage and spectral signals to estimate semantic alphabet size from few LLM samples, yielding better performance than baselines in low-sample regimes for alphabet estimation and QA error detection.
-
Complementing Self-Consistency with Cross-Model Disagreement for Uncertainty Quantification
Cross-model semantic disagreement adds an epistemic uncertainty term that improves total uncertainty estimation over self-consistency alone, helping flag confident errors in LLMs.
-
Do Hallucination Neurons Generalize? Evidence from Cross-Domain Transfer in LLMs
Hallucination neurons in LLMs are domain-specific, with cross-domain classifiers dropping from AUROC 0.783 within-domain to 0.563 across domains.
-
High-Entropy Tokens as Multimodal Failure Points in Vision-Language Models
Attacking only the top 20% high-entropy token positions in vision-language models causes comparable semantic damage and more harmful outputs than global attacks, and these vulnerable tokens transfer across model architectures.
-
Clotho: Measuring Task-Specific Pre-Generation Test Adequacy for LLM Inputs
Clotho ranks LLM test inputs by failure likelihood using a Gaussian mixture model over pre-generation hidden states of passing inputs, reaching 0.716 ROC-AUC with 5.4% labelling and transferring from open-weight to proprietary LLMs.
-
GrACE: A Generative Approach to Better Confidence Elicitation and Efficient Test-Time Scaling in Large Language Models
GrACE is a fine-tuned generative method that uses similarity to a special token embedding for real-time calibrated confidence in LLMs and enables efficient confidence-based test-time scaling.
-
The Origins of Stochasticity: Comprehensive Investigations on Uncertainty Quantification for Large Language Models
The paper introduces a four-source uncertainty taxonomy for LLMs and finds that consensus-based UQ methods outperform others while larger models show lower uncertainty estimates.
-
FLaG: Fine-Grained Latent Grouping for Hallucination Detection
FLaG models hallucination detection via latent evidence groups and energy-based routing with log-marginal aggregation, claiming SOTA results and a theoretical link to Bayes-optimal detection under heterogeneous mechanisms.
-
Capability Self-Assessment: Teaching LLMs to Know Their Limits
Reinforcement learning teaches LLMs to assess their own capabilities more effectively than supervised fine-tuning, preserves original skills, generalizes out of distribution, and aids local-cloud routing and data selection.
-
Reverse Probing: Supervised Token-level Uncertainty Quantification for Large Language Models in Clinical Text
Reverse Probing extracts token-level uncertainty from LLM internal activations on labeled clinical summaries, outperforming eight baselines with up to 4x higher AUPRC on two expert-annotated datasets while lowering compute costs.
-
MultiHaluDet: Multilingual Hallucination Detection via LLM Hidden State Probing
MultiHaluDet uses multi-layer hidden-state probing, multi-scale attention, and a calibrated classifier ensemble to detect multilingual hallucinations, reporting up to 98.55% AUROC on English benchmarks and strong cross-lingual transfer to French, Bangla, and Amharic.
-
LATERN: Test-Time Context-Aware Explainable Video Anomaly Detection
LATERN reformulates video anomaly detection as temporal evidence aggregation via context-aware scoring (CEA) and recursive aggregation (REA) to improve accuracy and coherence for frozen VLMs on benchmarks like UCF-Crime.
-
Chain-of-Thought as a Lens: Evaluating Structured Reasoning Alignment between Human Preferences and Large Language Models
The Alignment Score quantifies semantic divergence between model-generated and human-preferred reasoning chains and correlates with accuracy, readability, and coherence.
-
Large Lemma Miners: Can LLMs do Induction Proofs for Hardware?
LLMs, verified by a symbolic model checker, produced correct inductive strengthenings for 82 of 94 curated RTL safety properties.
-
TokUR: Token-Level Uncertainty Estimation for Large Language Model Reasoning
TokUR estimates token-level uncertainty via low-rank weight perturbations in LLMs, aggregates signals to correlate with correctness, and uses them to improve reasoning performance on math tasks.
-
Rethinking Uncertainty Estimation in LLMs: A Principled Single-Sequence Measure
Negative log-likelihood of the greedy-decoded most likely sequence (G-NLL) is a principled single-sequence uncertainty measure for LLMs that achieves state-of-the-art results.
-
Online Safety Monitoring for LLMs
Simple thresholding on an external verifier signal, calibrated by risk control, performs competitively with sequential hypothesis testing monitors on math reasoning and red-teaming datasets.
- To Call or Not to Call: A Framework to Assess and Optimize LLM Tool Calling
- Process Supervision of Confidence Margin for Calibrated LLM Reasoning
- Convergent Evolution: How Different Language Models Learn Similar Number Representations