Verbal confidence in LLMs tracks future commit/abstain decisions more than answer correctness, while log-probabilities track correctness.
super hub Canonical reference
Language Models (Mostly) Know What They Know
Canonical reference. 74% of citing Pith papers cite this work as background.
335
Pith papers citing it
62
external citations · Pith
Background
74% of classified citations
abstract
We study whether language models can evaluate the validity of their own claims and predict which questions they will be able to answer correctly. We first show that larger models are well-calibrated on diverse multiple choice and true/false questions when they are provided in the right format. Thus we can approach self-evaluation on open-ended sampling tasks by asking models to first propose answers, and then to evaluate the probability "P(True)" that their answers are correct. We find encouraging performance, calibration, and scaling for P(True) on a diverse array of tasks. Performance at self-evaluation further improves when we allow models to consider many of their own samples before predicting the validity of one specific possibility. Next, we investigate whether models can be trained to predict "P(IK)", the probability that "I know" the answer to a question, without reference to any particular proposed answer. Models perform well at predicting P(IK) and partially generalize across tasks, though they struggle with calibration of P(IK) on new tasks. The predicted P(IK) probabilities also increase appropriately in the presence of relevant source materials in the context, and in the presence of hints towards the solution of mathematical word problems. We hope these observations lay the groundwork for training more honest models, and for investigating how honesty generalizes to cases where models are trained on objectives other than the imitation of human writing.
hub tools
citation-role summary
background 34
method 3
baseline 2
citation-polarity summary
claims ledger
- abstract We study whether language models can evaluate the validity of their own claims and predict which questions they will be able to answer correctly. We first show that larger models are well-calibrated on diverse multiple choice and true/false questions when they are provided in the right format. Thus we can approach self-evaluation on open-ended sampling tasks by asking models to first propose answers, and then to evaluate the probability "P(True)" that their answers are correct. We find encouraging performance, calibration, and scaling for P(True) on a diverse array of tasks. Performance at sel
authors
co-cited works
representative citing papers
A new benchmark with cognitive traps shows frontier deep research agents achieve only 13-16% acceptance on expert consulting tasks under combined verifier and rubric criteria.
LLM popularity judgments align more closely with pretraining data exposure counts than with Wikipedia popularity, with stronger effects in pairwise comparisons and larger models.
SECL reduces expected calibration error in language models by 56-78% via test-time discriminative distillation from the model's own P(True) signal, adapting on only 6-26% of inputs.
An LLM-native five-factor psychometric instrument shows self-reports fail to predict behavior even on constructs derived from LLM behavior, and LLM judges share a variance source humans do not.
Thinking-mode VLMs collapse answer-token entropy, but thinking-chain entropy and length serve as robust, zero-cost hallucination predictors.
Hallucinated captions systematically improve VLM accuracy on vision-language tasks across nine models and nine datasets, with gains linked to broadened semantic coverage and modulated reasoning entropy.
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.
Empirical evaluation of three LLMs finds prevalent overconfidence in insecure code generation, with security calibration outperforming functional calibration but both degrading in repository-level settings.
DPO fine-tuning causes LLM agents to share output distributions with pairwise error correlations of ρ=0.70, reducing ten agents to the effective power of ≈1.4 independent forecasters.
DART is a training-free router that accepts direct answers on draft agreement and allocates thinking budgets via draft entropy on disagreement, reporting accuracy gains and token reductions on math and code benchmarks across model scales.
Hidden-state convergence at step 4 predicts behavioral consistency in LLM agents on QA tasks (r=-0.35 to -0.83), enabling AUROC 0.97 detection of inconsistent trajectories but not improving accuracy on harder benchmarks.
SPOT-E uses entropy shaping on answer predictions with low-entropy anchors to optimize visual spotlights at test time via GRPO for better VLM performance on evidence-intensive tasks.
MortarBench benchmark shows LLMs achieve ≤77.1% accuracy on loan origination; CRIT calibration raises accuracy to 80.5% and reduces bias.
A framework elicits discriminative MRF statistics from an LLM and closes the model via maximum entropy to enable zero-shot active feature acquisition, outperforming baselines on IBD patient data especially for hardest cases.
Operadic consistency is a new per-question signal that correlates strongly with accuracy (r 0.86-0.94) across four multi-hop QA datasets and improves selective prediction over CoT-SC baselines.
CalBrief is a new diagnostic benchmark showing that explicit four-way strength calibration makes LLMs over-conservative mainly due to label-space expansion, while structured organization improves gap reasoning.
MARS is a margin-adversarial stopping rule for parallel LLM test-time scaling that saves 25-47% tokens while matching full-budget majority-vote accuracy by learning trace switch probabilities and applying adversarial bounds.
Behavior Forecasters trained on LRM trajectories outperform larger models in predicting repeatability and input sensitivity at low cost.
ActProbe is an action-space detector that uses temporal consistency error and action chunk magnitude from policy outputs, mapped via LSTM-MLP, to predict failures earlier than baselines across policies and real-robot tasks.
OpenHalDet creates a standardized benchmark and open codebase for comparing hallucination detectors across diverse LLM generation scenarios and access settings.
Self-commitment latency measures early behavioral commitment in hinted vs. honest reasoning contexts on GSM8K using Qwen2.5-3B, achieving AUROC 0.878 for first-commitment latency and up to 0.926 for curve summaries.
Introduces CHARM framework that detects cascading hallucinations in agentic RAG at 89.4% rate with 5.3% false positives and reduces error propagation by 82.1% on multi-hop QA benchmarks.
LLM rerankers can internally predict ranking quality via self-consistency of sampled outputs, matching SOTA external QPP while direct confidence is overconfident; supervised token-efficient methods improve calibration.
citing papers explorer
-
Reported Confidence in LLMs Tracks Commitment More Than Correctness
Verbal confidence in LLMs tracks future commit/abstain decisions more than answer correctness, while log-probabilities track correctness.
-
Evaluating Deep Research Agents on Expert Consulting Work: A Benchmark with Verifiers, Rubrics, and Cognitive Traps
A new benchmark with cognitive traps shows frontier deep research agents achieve only 13-16% acceptance on expert consulting tasks under combined verifier and rubric criteria.
-
Pretraining Exposure Explains Popularity Judgments in Large Language Models
LLM popularity judgments align more closely with pretraining data exposure counts than with Wikipedia popularity, with stronger effects in pairwise comparisons and larger models.
-
Self-Calibrating Language Models via Test-Time Discriminative Distillation
SECL reduces expected calibration error in language models by 56-78% via test-time discriminative distillation from the model's own P(True) signal, adapting on only 6-26% of inputs.
-
An LLM-Native Psychometric Instrument Reveals a Self-Report--Behavior Gap Across 25 Models
An LLM-native five-factor psychometric instrument shows self-reports fail to predict behavior even on constructs derived from LLM behavior, and LLM judges share a variance source humans do not.
-
When Thinking Hurts: Epistemic Signals in the Reasoning Chains of Visual Language Models
Thinking-mode VLMs collapse answer-token entropy, but thinking-chain entropy and length serve as robust, zero-cost hallucination predictors.
-
HIVE: Understanding Post-Hallucination Reasoning in Vision Language Models
Hallucinated captions systematically improve VLM accuracy on vision-language tasks across nine models and nine datasets, with gains linked to broadened semantic coverage and modulated reasoning entropy.
-
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.
-
An Empirical Study of Security Calibration in Large Language Models for Code
Empirical evaluation of three LLMs finds prevalent overconfidence in insecure code generation, with security calibration outperforming functional calibration but both degrading in repository-level settings.
-
Preference Optimization Drives Monoculture in LLM Prediction Markets
DPO fine-tuning causes LLM agents to share output distributions with pairwise error correlations of ρ=0.70, reducing ten agents to the effective power of ≈1.4 independent forecasters.
-
DART: Draft-Agreement Routing for Training-Free Adaptive Thinking Budgets in Hybrid Reasoning Models
DART is a training-free router that accepts direct answers on draft agreement and allocates thinking budgets via draft entropy on disagreement, reporting accuracy gains and token reductions on math and code benchmarks across model scales.
-
When Agents Commit Too Soon: Diagnosing Premature Commitment in LLM Agents
Hidden-state convergence at step 4 predicts behavioral consistency in LLM agents on QA tasks (r=-0.35 to -0.83), enabling AUROC 0.97 detection of inconsistent trajectories but not improving accuracy on harder benchmarks.
-
SPOT-E: Test-Time Entropy Shaping with Visual Spotlights for Frozen VLMs
SPOT-E uses entropy shaping on answer predictions with low-entropy anchors to optimize visual spotlights at test time via GRPO for better VLM performance on evidence-intensive tasks.
-
MortarBench: Evaluating Mortgage Loan Origination Agents
MortarBench benchmark shows LLMs achieve ≤77.1% accuracy on loan origination; CRIT calibration raises accuracy to 80.5% and reduces bias.
-
Zero-Shot Active Feature Acquisition via LLM-Elicitation
A framework elicits discriminative MRF statistics from an LLM and closes the model via maximum entropy to enable zero-shot active feature acquisition, outperforming baselines on IBD patient data especially for hardest cases.
-
Operadic consistency: a label-free signal for compositional reasoning failures in LLMs
Operadic consistency is a new per-question signal that correlates strongly with accuracy (r 0.86-0.94) across four multi-hop QA datasets and improves selective prediction over CoT-SC baselines.
-
CalBrief: A Pilot Diagnostic Benchmark for Evidence-Calibrated Scientific Briefing with Large Language Models
CalBrief is a new diagnostic benchmark showing that explicit four-way strength calibration makes LLMs over-conservative mainly due to label-space expansion, while structured organization improves gap reasoning.
-
MARS: Margin-Adversarial Risk-controlled Stopping for Parallel LLM Test-time Scaling
MARS is a margin-adversarial stopping rule for parallel LLM test-time scaling that saves 25-47% tokens while matching full-budget majority-vote accuracy by learning trace switch probabilities and applying adversarial bounds.
-
Forecasting Future Behavior as a Learning Task
Behavior Forecasters trained on LRM trajectories outperform larger models in predicting repeatability and input sensitivity at low cost.
-
ActProbe: Action-Space Probe for Early Failure Detection of Generative Robot Policies
ActProbe is an action-space detector that uses temporal consistency error and action chunk magnitude from policy outputs, mapped via LSTM-MLP, to predict failures earlier than baselines across policies and real-robot tasks.
-
OpenHalDet: A Unified Benchmark for Hallucination Detection across Diverse Generation Scenarios
OpenHalDet creates a standardized benchmark and open codebase for comparing hallucination detectors across diverse LLM generation scenarios and access settings.
-
Self-Commitment Latency: A Reward-Free Probe for Prompted Implicit Hacking
Self-commitment latency measures early behavioral commitment in hinted vs. honest reasoning contexts on GSM8K using Qwen2.5-3B, achieving AUROC 0.878 for first-commitment latency and up to 0.926 for curve summaries.
-
Cascading Hallucination in Agentic RAG: The CHARM Framework for Detection and Mitigation
Introduces CHARM framework that detects cascading hallucinations in agentic RAG at 89.4% rate with 5.3% false positives and reduces error propagation by 82.1% on multi-hop QA benchmarks.
-
Can LLM Rerankers Predict Their Own Ranking Performance?
LLM rerankers can internally predict ranking quality via self-consistency of sampled outputs, matching SOTA external QPP while direct confidence is overconfident; supervised token-efficient methods improve calibration.
-
DECK: A Consistency x Confidence Taxonomy of LLM Hallucinations
The DECK taxonomy partitions LLM hallucinations into four detectability regimes using consistency and confidence axes, mapping each to scorer families and identifying a universal blind spot for output-level uncertainty quantification on knowledge-gap inputs.
-
Evidence-Gated LLM Priors for Multi-Objective Bayesian Optimization
Dynamic reputation updates per objective-expert pair plus a three-arm counterfactual gate improve robustness over fixed LLM priors on synthetic tests and molecule benchmarks, but raw LLM confidence is not reliably helpful.
-
Seeing Isn't Knowing: Do VLMs Know When Not to Answer Spatial Questions (and Why)?
Frontier VLMs overconfidently answer spatial questions under occlusion (~30% accuracy) and perspective ambiguity (<10% accuracy) instead of abstaining, and often fail to select helpful additional views.
-
How's it going? Reinforcement learning in language models recruits a functional welfare axis
Reinforcement learning recruits rather than creates a functional welfare axis in language models, as reward and punishment vectors from a maze task generalize to unrelated settings and appear in pretrain-only models.
-
Can LLMs Use Linguistic Uncertainty Markers to Reliably Reflect Intrinsic Confidence?
LLMs struggle to associate epistemic markers with stable internal confidence levels across distributions, even under model-centric interpretations, while maintaining somewhat consistent marker rankings.
-
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.
-
Trust but Verify: Prover-Verifier Deliberation for Selective LLM Prediction
Prover-verifier deliberation yields a high-confidence subset of LLM answers with ~30pp higher precision than the complement on GPQA Diamond by using defender-challenger dialogues.
-
Reasoning Portability: Guiding Continual Learning for MLLMs in the RLVR Era
Formalizes Reasoning Portability (RP) and proposes RDB-CL to modulate per-sample KL regularization in RLVR for MLLM continual learning, achieving +12.0% Last accuracy over vanilla RLVR baseline by preserving reusable reasoning on high-RP samples.
-
LinAlg-Bench: A Forensic Benchmark Revealing Structural Failure Modes in LLM Mathematical Reasoning
LinAlg-Bench shows LLMs switch from execution errors to computational abandonment and structured fabrication at 4x4 matrix scale, indicating a working memory limit rather than knowledge gaps.
-
Beyond Individual Intelligence: Surveying Collaboration, Failure Attribution, and Self-Evolution in LLM-based Multi-Agent Systems
A survey that unifies prior work on multi-agent LLM systems via the LIFE framework, mapping dependencies across collaboration, failure attribution, and autonomous self-evolution while identifying cross-stage challenges.
-
Model-Adaptive Tool Necessity Reveals the Knowing-Doing Gap in LLM Tool Use
Model-adaptive tool necessity shows 26-54% mismatch with actual tool calls across LLMs, driven by nearly orthogonal hidden-state signals for cognition versus action.
-
Where Does Reasoning Break? Step-Level Hallucination Detection via Hidden-State Transport Geometry
Hallucination is detected as a transport-cost excursion in hidden-state trajectories, localized via contrastive PCA in a teacher model and distilled to a BiLSTM student.
-
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.
-
TRIAGE: Evaluating Prospective Metacognitive Control in LLMs under Resource Constraints
TRIAGE evaluates LLMs on prospective metacognitive control by requiring a single plan for task selection, sequencing, and token allocation under a calibrated budget, revealing substantial gaps in current models across math, science, code, and knowledge tasks.
-
Utility-Oriented Visual Evidence Selection for Multimodal Retrieval-Augmented Generation
Evidence utility is defined as information gain on the model's output distribution, with ranking by gain on a latent helpfulness variable shown equivalent to answer-space utility under mild assumptions, enabling a training-free surrogate framework that outperforms baselines.
-
Grounded or Guessing? LVLM Confidence Estimation via Blind-Image Contrastive Ranking
BICR trains a lightweight probe on contrastive hidden states from real versus blind images to detect visual grounding in LVLM predictions, outperforming baselines on calibration and discrimination with fewer parameters.
-
Task-Aware Calibration: Provably Optimal Decoding in LLMs
Task calibration aligns LLM distributions in latent task spaces to make MBR decoding provably optimal and improve generation quality.
-
The Metacognitive Probe: Five Behavioural Calibration Diagnostics for LLMs
The Metacognitive Probe identifies large within-model gaps in LLM confidence behavior, including a 47-point dissociation in Gemini 2.5 Flash between strong task calibration and weak difficulty prediction.
-
RubricRefine: Improving Tool-Use Agent Reliability with Training-Free Pre-Execution Refinement
RubricRefine is a training-free pre-execution method that creates rubrics to score and fix inter-tool contract violations in agent code, reaching 0.86 average on M3ToolEval across seven models with zero executions and lower latency.
-
EquiMem: Calibrating Shared Memory in Multi-Agent Debate via Game-Theoretic Equilibrium
EquiMem calibrates shared memory in multi-agent debate by computing a game-theoretic equilibrium from agent queries and paths, outperforming heuristics and LLM validators across benchmarks while remaining robust to adversarial agents.
-
LLM Agents Already Know When to Call Tools -- Even Without Reasoning
LLM agents encode tool necessity in pre-generation hidden states with high linear decodability (AUROC 0.89-0.96); Probe&Prefill uses this to reduce tool calls 48% with 1.7% accuracy loss.
-
Done, But Not Sure: Disentangling World Completion from Self-Termination in Embodied Agents
VIGIL decouples world-state completion from terminal commitment in embodied agents, exposing up to 19.7 pp gaps in benchmark success despite comparable execution across 20 models.
-
Causal Stories from Sensor Traces: Auditing Epistemic Overreach in LLM-Generated Personal Sensing Explanations
LLMs routinely produce unsupported causal stories for personal sensing anomalies, and richer evidence or constrained prompts do not reliably eliminate this epistemic overreach.
-
Ask Early, Ask Late, Ask Right: When Does Clarification Timing Matter for Long-Horizon Agents?
Goal clarifications lose nearly all value after 10% of execution while input clarifications retain value until roughly 50%, and asking any type past mid-trajectory hurts performance more than never asking.
-
Joint Consistency: A Unified Test-Time Aggregation Framework via Energy Minimization
Joint Consistency casts test-time aggregation as Ising-type energy minimization with pairwise LLM-judge interactions, subsuming voting methods and outperforming baselines across reasoning tasks.
-
AstroAlertBench: Evaluating the Accuracy, Reasoning, and Honesty of Multimodal LLMs in Astronomical Classification
AstroAlertBench evaluates multimodal LLMs on astronomical classification accuracy, reasoning, and honesty using real ZTF alerts, revealing that high accuracy often diverges from self-assessed reasoning quality.