REVIEW 11 cited by
Uncertainty-Based Abstention in LLMs Improves Safety and Reduces Hallucinations
Not yet reviewed by Pith; the record is open.
This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.
SPECIMEN: schema-true, not a live event
T0 review · schema-true
One-sentence machine reading of the paper's core claim.
pith:XXXXXXXX · record.json · timestamp
Uncertainty-Based Abstention in LLMs Improves Safety and Reduces Hallucinations
read the original abstract
A major barrier towards the practical deployment of large language models (LLMs) is their lack of reliability. Three situations where this is particularly apparent are correctness, hallucinations when given unanswerable questions, and safety. In all three cases, models should ideally abstain from responding, much like humans, whose ability to understand uncertainty makes us refrain from answering questions we don't know. Inspired by analogous approaches in classification, this study explores the feasibility and efficacy of abstaining while uncertain in the context of LLMs within the domain of question-answering. We investigate two kinds of uncertainties, statistical uncertainty metrics and a distinct verbalized measure, termed as In-Dialogue Uncertainty (InDU). Using these uncertainty measures combined with models with and without Reinforcement Learning with Human Feedback (RLHF), we show that in all three situations, abstention based on the right kind of uncertainty measure can boost the reliability of LLMs. By sacrificing only a few highly uncertain samples we can improve correctness by 2% to 8%, avoid 50% hallucinations via correctly identifying unanswerable questions and increase safety by 70% up to 99% with almost no additional computational overhead.
Forward citations
Cited by 11 Pith papers
-
Reflect-R1: Evidence-Driven Reflection for Self-Correction in Long Video Understanding
Reflect-R1 introduces the first evidence-driven self-correction framework for long video understanding using a three-stage pipeline, stage-decoupled RL via SD-GRPO, and a 120K dataset to achieve SOTA on VideoMME and L...
-
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 ad...
-
Evaluating medical AI under missing information: same-provider judges and human raters change apparent safety
In open-ended medical conversations with missing information, LLM judges are more lenient than clinicians and a model's same-provider judge can skew apparent safety rankings.
-
Reflect-R1: Evidence-Driven Reflection for Self-Correction in Long Video Understanding
Reflect-R1 introduces an evidence-driven intuition-verification-arbitration pipeline plus stage-decoupled RL (SD-GRPO) and a 120K dataset to enable grounded self-correction in long video models, claiming SOTA results.
-
Reflect-R1: Evidence-Driven Reflection for Self-Correction in Long Video Understanding
Reflect-R1 proposes an evidence-retrieval three-stage pipeline with SD-GRPO reinforcement learning and a new 120K dataset that claims SOTA results on VideoMME and LongVideoBench by grounding self-correction in objecti...
-
When Confidence Takes the Wrong Path: Diagnosing Retrieval-State Lock-In in RAG
Retrieval-state lock-in causes zero-dispersion errors in 42% of KG-RAG and 59% of dense-retrieval failures; a three-object check rule reaches 91.9% pooled precision at 7.7% coverage.
-
CRAFT: Cost-aware Refinement And Front-aware Tuning of Prompts
CRAFT is a Pareto-front prompt optimizer that allocates scarce LLM validation calls to candidates near the current front using accuracy- and cost-oriented generators plus NSGA-II retention.
-
No-Worse Context-Aware Decoding: Preventing Neutral Regression in Context-Conditioned Generation
NWCAD uses a two-stream setup with a two-stage gate to prevent accuracy drops on baseline-correct items under non-informative contexts while retaining gains from helpful contexts.
-
Causal Evidence that Language Models use Confidence to Drive Behavior
Language models deploy multidimensional internal confidence representations and threshold-based policies to control abstention behavior, with causal support from activation steering experiments.
-
Verbalizing LLM's Higher-order Uncertainty via Imprecise Probabilities
Prompting LLMs to report imprecise probability intervals (lower/upper confidence) instead of a single point value yields higher-order uncertainty scores that track prediction error and question ambiguity more coherent...
-
Steering the Verifiability of Multimodal AI Hallucinations
Researchers create a human-labeled dataset of obvious and elusive multimodal hallucinations and use learned activation-space probes to control their verifiability in MLLMs.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.