LLM agents overcommit on non-complete tasks at 41.7% unless given explicit support-state categories, which raise typed deferral accuracy to 91.7%.
MASH: Modeling Abstention via Selective Help-Seeking
5 Pith papers cite this work. Polarity classification is still indexing.
5
Pith papers citing it
abstract
LLMs cannot reliably recognize their parametric knowledge boundaries and often hallucinate answers to outside-of-boundary questions. In this paper, we introduce MASH (Modeling Abstention via Selective Help-seeking), a training framework that readily extracts abstentions from LLMs. Our key idea is that any external help-seeking by an LLM, i.e. search tool use, can serve as a proxy for abstention if the external help (search) is appropriately penalized while also rewarding answer accuracy. MASH operationalizes this idea using reinforcement learning with a pay-per-search reward. We run experiments on three knowledge-intensive QA datasets. Our results show that MASH substantially improves upon the selective help-seeking performance of prior efficient search approaches; on multi-hop datasets, it improves answer accuracy by 7.6%. Furthermore, MASH demonstrates strong off-the-shelf abstention performance, showcasing behavior competitive with prior abstention methods that additionally require predetermining model knowledge boundaries to construct training data. Overall, we show MASH training effectively aligns search tool use with parametric knowledge, which can be successfully leveraged for making abstention decisions and efficient search tool use
years
2026 5verdicts
UNVERDICTED 5representative citing papers
CPT is introduced as a pairwise reasoning-trace comparison stage that improves the reasoning-metacognition trade-off over standard SFT+RL pipelines across model scales.
TIAR uses trajectory-informed advantage reweighting during GRPO to improve LLM abstention F1 scores on AbstentionBench while preserving accuracy.
KARL uses a knowledge-boundary-aware reward from within-group response statistics and two-stage RL training to align LLM abstention with actual knowledge, yielding a better accuracy-hallucination trade-off on benchmarks.
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.
citing papers explorer
-
Don't Start What You Can't Finish: A Counterfactual Audit of Support-State Triage in LLM Agents
LLM agents overcommit on non-complete tasks at 41.7% unless given explicit support-state categories, which raise typed deferral accuracy to 91.7%.
-
Enhancing LLM Metacognition via Cognitive Pairwise Training
CPT is introduced as a pairwise reasoning-trace comparison stage that improves the reasoning-metacognition trade-off over standard SFT+RL pipelines across model scales.
-
TIAR: Trajectory-Informed Advantage Reweighting for LLM Abstention Learning
TIAR uses trajectory-informed advantage reweighting during GRPO to improve LLM abstention F1 scores on AbstentionBench while preserving accuracy.
-
KARL: Mitigating Hallucinations in LLMs via Knowledge-Boundary-Aware Reinforcement Learning
KARL uses a knowledge-boundary-aware reward from within-group response statistics and two-stage RL training to align LLM abstention with actual knowledge, yielding a better accuracy-hallucination trade-off on benchmarks.
-
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.