Research map / Large language models / Retrieval and knowledge
MASH: Modeling Abstention via Selective Help-Seeking
Mustafa Omer Gul, Claire Cardie, Tanya Goyal
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add warm-start SFT with synthetic diverse-search trajectories (ablation) | MASH w/ OTC-Strict · HotPotQA | accuracy: 26.97 -> 53.32 | Improved |
| add warm-start SFT with synthetic diverse-search trajectories (ablation) | MASH w/ OTC-Strict · 2Wiki | accuracy: 10.41 -> 46.23 | Improved |
| add warm-start SFT with synthetic diverse-search trajectories (ablation) | MASH w/ Exp · 2Wiki | accuracy: 9.72 -> 44.32 | Improved |
| add warm-start SFT with synthetic diverse-search trajectories (ablation) | OTC reward · HotPotQA | accuracy: 44.76 -> 55.42 | Improved |
| add warm-start SFT with synthetic diverse-search trajectories (ablation) | OTC reward · 2Wiki | accuracy: 39.59 -> 45.99 | Improved |
| add warm-start SFT with synthetic diverse-search trajectories (ablation) | MASH w/ OTC-Strict · Natural Questions | accuracy: 52.35 -> 56.37 | Mixed |
| use strict penalty OTC-Strict instead of OTC | MASH · Natural Questions | tool productivity: 29.95 -> 38.63 | Improved |
| use strict penalty OTC-Strict instead of OTC | MASH · Natural Questions | accuracy: 59.81 -> 56.37 | Hurt |
| use strict penalty OTC-Strict instead of OTC | MASH · 2Wiki | tool productivity: 18.87 -> 19.08 | No clear effect |
| use MASH w/ OTC-Strict instead of baseline OTC | OTC · HotPotQA | accuracy: 44.76 -> 53.32 | Improved |
| use MASH w/ OTC-Strict instead of baseline OTC | OTC · Natural Questions | tool productivity: 29.45 -> 38.63 | Improved |
| replace DPO abstention training with MASH w/ OTC-Strict | DPO · HotPotQA | precision on non-abstained: 53.1 -> 59.9 | Improved |
Closest papers by meaning
- AbstentionBench: Reasoning LLMs Fail on Unanswerable Questions (NeurIPS 2025 · similarity 0.87)
- ReSearch: Learning to Reason with Search for LLMs via Reinforcement Learning (NeurIPS 2025 · similarity 0.86)
- AceSearcher: Bootstrapping Reasoning and Search for LLMs via Reinforced Self-Play (NeurIPS 2025 · similarity 0.86)
- Agentic Abstention: Do Agents Know When to Stop Instead of Act? (NeurIPS 2026 · similarity 0.85)
- SD-Search: Hindsight Self-Distillation for Search-Augmented Reasoning (NeurIPS 2026 · similarity 0.85)
- Explore What LLM Does Not Know in Complex Question Answering (AAAI 2025 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
