AI Paper Atlas

Research map / Large language models / Retrieval and knowledge

MASH: Modeling Abstention via Selective Help-Seeking

Mustafa Omer Gul, Claire Cardie, Tanya Goyal

ICML 2026 · Large language models › Retrieval and knowledge · approach: Explanatory and multiple-choice QA benchmarks · 0 citations (OpenAlex)

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 changedWhereMeasuredEffect
add warm-start SFT with synthetic diverse-search trajectories (ablation)MASH w/ OTC-Strict · HotPotQAaccuracy: 26.97 -> 53.32Improved
add warm-start SFT with synthetic diverse-search trajectories (ablation)MASH w/ OTC-Strict · 2Wikiaccuracy: 10.41 -> 46.23Improved
add warm-start SFT with synthetic diverse-search trajectories (ablation)MASH w/ Exp · 2Wikiaccuracy: 9.72 -> 44.32Improved
add warm-start SFT with synthetic diverse-search trajectories (ablation)OTC reward · HotPotQAaccuracy: 44.76 -> 55.42Improved
add warm-start SFT with synthetic diverse-search trajectories (ablation)OTC reward · 2Wikiaccuracy: 39.59 -> 45.99Improved
add warm-start SFT with synthetic diverse-search trajectories (ablation)MASH w/ OTC-Strict · Natural Questionsaccuracy: 52.35 -> 56.37Mixed
use strict penalty OTC-Strict instead of OTCMASH · Natural Questionstool productivity: 29.95 -> 38.63Improved
use strict penalty OTC-Strict instead of OTCMASH · Natural Questionsaccuracy: 59.81 -> 56.37Hurt
use strict penalty OTC-Strict instead of OTCMASH · 2Wikitool productivity: 18.87 -> 19.08No clear effect
use MASH w/ OTC-Strict instead of baseline OTCOTC · HotPotQAaccuracy: 44.76 -> 53.32Improved
use MASH w/ OTC-Strict instead of baseline OTCOTC · Natural Questionstool productivity: 29.45 -> 38.63Improved
replace DPO abstention training with MASH w/ OTC-StrictDPO · HotPotQAprecision on non-abstained: 53.1 -> 59.9Improved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy