AI Paper Atlas

Research map / Large language models / Retrieval and knowledge

Train for Truth, Keep the Skills: Binary Retrieval-Augmented Reward Mitigates Hallucinations

Tong Chen, Akari Asai, Luke Zettlemoyer, Hannaneh Hajishirzi, Faeze Brahman

ICML 2026 · Large language models › Retrieval and knowledge · approach: Retrieval-augmented reasoning · 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
train with binary retrieval-augmented reward instead of continuous VeriScore reward (ablation)Qwen3-8B GRPO · ALPACAEVAL (length-controlled)win rate: 42.2 -> 53.9Improved
use binary RAR reward instead of VeriScore reward (ablation)Qwen3-8B GRPO · Short-form average (POPQA, GPQA)hallucination rate: 42.3 -> 27.6Improved
online RL with Binary RAR on top of base modelQwen3-8B · Long-form averagehallucination rate: 61.9 -> 37.5Improved
add SFT on most factual responseQwen3-8B · Long-form averagehallucination rate: 61.9 -> 60.9No clear effect
add DPO on factuality preference pairsQwen3-8B · Long-form averagehallucination rate: 61.9 -> 53.4Improved
use general LM-judge reward for RLQwen3-8B · Long-form averagehallucination rate: 61.9 -> 65.4Hurt
use Binary RAR with Qwen3-4B baseQwen3-4B · Short-form average (POPQA, GPQA)hallucination rate: 68.7 -> 41.9Improved
apply Binary RAR to a different model familyTulu3-8B · BIOGRAPHYhallucination rate: 68.1 -> 60.1Improved
apply Binary RAR to Tulu3-8BTulu3-8B · ALPACAEVALwin rate: 24.3 -> 23.4No clear effect
evaluate POPQA and GPQA with no abstention after Binary RARQwen3-8B · POPQA, GPQA (no abstention)accuracy: POPQA 20.2 -> 20.6; GPQA 48.2 -> 48.8No clear effect
threshold VeriScore at 0.5 as a binary reward (claim-level) instead of response-level binary RAR (ablation)Qwen3-8B RL · ALPACAEVAL vs BIOGRAPHY FactScorehallucination-utility trade-offHurt
use conflict-only claim reward (fraction of non-contradictory claims) instead of binary RAR (ablation)Qwen3-8B RL · ALPACAEVALwin rateHurt

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