A Controlled Study of Decoding-Time Truthfulness Methods on Instruction-Tuned LLMs

• URL: http://arxiv.org/abs/2606.12160v2
• Date: Thu, 11 Jun 2026 13:44:29 GMT
• Title: A Controlled Study of Decoding-Time Truthfulness Methods on Instruction-Tuned LLMs
• Abstract summary: Layer-contrast decoding, inference-time intervention, and learned logit adapters have demonstrated 10-30 point gains on TruthfulQA.<n>Modern instruction-tuned LLMs already achieve substantially higher baselines.<n> Deliberative prompting methods appear more robust in the evaluated regime.
• Score: 3.4007995136788
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Decoding-time truthfulness methods -- layer-contrast decoding, inference-time intervention, and learned logit adapters -- have demonstrated 10-30 point gains on TruthfulQA when applied to base language models. However, modern instruction-tuned LLMs already achieve substantially higher baselines (61-76%), raising the question of whether these methods remain effective in practice. We design a six-control evaluation framework -- out-of-distribution training, multi-judge validation, simple decoding baselines, confound controls, bootstrap confidence intervals, and seed variance -- and apply it across 5 models (1B-70B), 3 benchmarks, and 15 methods. We find that previously reported gains shrink substantially under strict controls: on the full TruthfulQA benchmark (N=817), no token-level method achieves statistically significant improvement, and the best learned adapter scores -2.0 points below greedy (p=.23). We identify five evaluation sensitivities -- contamination, judge choice, missing baselines, confounds, and statistical noise -- that individually or jointly account for these discrepancies. Cross-benchmark validation on HaluEval QA and TriviaQA confirms that these patterns extend beyond TruthfulQA. Deliberative prompting methods (chain-of-thought, self-critique) appear more robust in the evaluated regime, with CoT achieving +5.6-19pp across benchmarks as a training-free, single-pass method. We release a seven-point evaluation checklist and discuss implications for future truthfulness research.

Related papers

• DataFlex-RL: An Evaluation Platform for RLVR Data Policies [11.073211317719393]
  Data policies for reinforcement learning with verifiable rewards determine which rollouts are used, how strongly they are weighted, and which domains contribute to subsequent training batches.<n>We introduce DataFlex-RL, an evaluation platform for comparing these choices under a common GRPO recipe.<n>Our primary experiment evaluates 13 configurations across 12 matched seeds using Qwen2.5-7B-Base and 12 mathematics, logic, and science benchmarks.
  arXiv  Detail & Related papers  (2026-09-05T14:04:50Z)
• FADE: From Passive Verification to Active Discovery in Counterfactual Video Understanding [19.559643217912114]
  Existing multiple-choice question (MCQ) benchmarks inadvertently leak target events through their questions and candidate options.<n>We present FADE, an effective training framework for counterfactual discovery and explanation.<n>FADE achieves state-of-the-art strict paired scores across all three tasks on DualityVidQA-test and IPV-Bench, outperforming GPT-5.6.
  arXiv  Detail & Related papers  (2026-08-11T10:22:08Z)
• A Nash Equilibrium Framework For Training-Free Multimodal Step Verification [36.36755437383068]
  We propose a training-free verification approach that treats step-wise verification as a coordination problem among specialized judges.<n>Our approach achieves consistent improvements of 2.4% to 5.2% over baseline models.
  arXiv  Detail & Related papers  (2026-05-19T15:54:22Z)
• Scalable Token-Level Hallucination Detection in Large Language Models [63.3426544914783]
  Internal hallucinations are difficult to detect in reasoning-intensive tasks.<n>We propose TokenHD, a holistic pipeline for training token-level hallucination detectors.
  arXiv  Detail & Related papers  (2026-05-12T16:47:40Z)
• Evaluating Prompt Injection Defenses for Educational LLM Tutors: Security-Usability-Latency Trade-offs [51.56484100374058]
  guardrail design entails explicit trade-offs among adversarial robustness, benign-task usability, and response latency.<n>We evaluate a domain-specific multi-layer safeguard pipeline combining deterministic pattern filters, structural validation, contextual sandboxing, and session-level behavioral checks.<n>Results expose operational trade-offs: NeMo reaches 0 percent bypass at 16.22 percent FPR and roughly 1.5s latency, while Prompt Guard yields 38.48 percent bypass with 3.60 percent FPR.
  arXiv  Detail & Related papers  (2026-03-29T18:52:01Z)
• Locate-then-Sparsify: Attribution Guided Sparse Strategy for Visual Hallucination Mitigation [68.41785694664011]
  We propose a plug-and-play framework called Locate-Then-Sparsify for Feature Steering (LTS-FS)<n>It controls the steering intensity according to the hallucination relevance of each layer.<n>Our framework effectively mitigates hallucination while preserving strong performance.
  arXiv  Detail & Related papers  (2026-03-17T09:16:50Z)
• PRIME: A Process-Outcome Alignment Benchmark for Verifiable Reasoning in Mathematics and Engineering [71.15346406323827]
  We introduce PRIME, a benchmark for evaluating verifiers on Process-Outcome Alignment verification.<n>We find that current verifiers frequently fail to detect derivation flaws.<n>We propose a process-aware RLVR training paradigm utilizing verifiers selected via PRIME.
  arXiv  Detail & Related papers  (2026-02-12T04:45:01Z)
• Confident Rankings with Fewer Items: Adaptive LLM Evaluation with Continuous Scores [25.638175689769934]
  We present a principled extension of IRT-based adaptive testing to continuous bounded scores (ROUGE, BLEU, LLM-as-a-Judge)<n>We introduce an uncertainty aware ranker with adaptive stopping criteria that achieves reliable model ranking while testing as few items as possible.<n>Our method uses 2% of the items while improving ranking correlation by 0.12 over random sampling, with 95% accuracy on confident predictions.
  arXiv  Detail & Related papers  (2026-01-20T11:59:13Z)
• Encyclo-K: Evaluating LLMs with Dynamically Composed Knowledge Statements [78.87065404966002]
  Existing benchmarks predominantly curate questions at the question level.<n>We propose Encyclo-K, a statement-based benchmark that rethinks benchmark construction from the ground up.
  arXiv  Detail & Related papers  (2025-12-31T13:55:54Z)
• Verifying Large Language Models' Reasoning Paths via Correlation Matrix Rank [71.09032766271493]
  Large language models (LLMs) are prone to errors and hallucinations.<n>How to check their outputs effectively and efficiently has become a critical problem in their applications.
  arXiv  Detail & Related papers  (2025-10-28T11:01:10Z)
• SHALE: A Scalable Benchmark for Fine-grained Hallucination Evaluation in LVLMs [52.03164192840023]
  Large Vision-Language Models (LVLMs) still suffer from hallucinations, i.e., generating content inconsistent with input or established world knowledge.<n>We propose an automated data construction pipeline that produces scalable, controllable, and diverse evaluation data.<n>We construct SHALE, a benchmark designed to assess both faithfulness and factuality hallucinations.
  arXiv  Detail & Related papers  (2025-08-13T07:58:01Z)
• ICR Probe: Tracking Hidden State Dynamics for Reliable Hallucination Detection in LLMs [50.18087419133284]
  hallucination detection methods leveraging hidden states predominantly focus on static and isolated representations.<n>We introduce a novel metric, the ICR Score, which quantifies the contribution of modules to the hidden states' update.<n>We propose a hallucination detection method, the ICR Probe, which captures the cross-layer evolution of hidden states.
  arXiv  Detail & Related papers  (2025-07-22T11:44:26Z)
• Robust Hallucination Detection in LLMs via Adaptive Token Selection [35.06045656558144]
  Hallucinations in large language models (LLMs) pose significant safety concerns that impede their broader deployment.<n>We propose HaMI, a novel approach that enables robust detection of hallucinations through adaptive selection and learning of critical tokens.<n>We achieve this robustness by an innovative formulation of the Hallucination detection task as Multiple Instance (HaMI) learning over token-level representations within a sequence.
  arXiv  Detail & Related papers  (2025-04-10T15:39:10Z)
• CHAIR -- Classifier of Hallucination as Improver [1.397828249435483]
  We introduce CHAIR (Classifier of Hallucination As ImproveR), a supervised framework for detecting hallucinations by analyzing internal logits from each layer of every token.<n>Our method extracts a compact set of features such as maximum, minimum, mean, standard deviation, and slope-from the token logits across all layers, enabling effective hallucination detection without overfitting.
  arXiv  Detail & Related papers  (2025-01-05T12:15:02Z)
• Improving LLM Reasoning through Scaling Inference Computation with Collaborative Verification [52.095460362197336]
  Large language models (LLMs) struggle with consistent and accurate reasoning. LLMs are trained primarily on correct solutions, reducing their ability to detect and learn from errors. We propose a novel collaborative method integrating Chain-of-Thought (CoT) and Program-of-Thought (PoT) solutions for verification.
  arXiv  Detail & Related papers  (2024-10-05T05:21:48Z)
• On Speeding Up Language Model Evaluation [48.51924035873411]
  We propose an $textitadaptive$ approach to explore this space.<n>We lean on multi-armed bandits to sequentially identify the next (method, validation sample)-pair to evaluate.<n>We show that it can identify the top-performing method using only 5-15% of the typical resources.
  arXiv  Detail & Related papers  (2024-07-08T17:48:42Z)
• A New Benchmark and Reverse Validation Method for Passage-level Hallucination Detection [63.56136319976554]
  Large Language Models (LLMs) generate hallucinations, which can cause significant damage when deployed for mission-critical tasks. We propose a self-check approach based on reverse validation to detect factual errors automatically in a zero-resource fashion. We empirically evaluate our method and existing zero-resource detection methods on two datasets.
  arXiv  Detail & Related papers  (2023-10-10T10:14:59Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.