AI Paper Atlas

Research map / Large language models / Alignment and preferences

Maximizing Mutual Information Between Prompt and Response Improves LLM Performance With No Additional Data

HyunJi Nam, Haoran Li, Natasha Jaques

ICML 2026 · Large language models › Alignment and preferences · approach: Active and token-level preference learning · 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 MIPO contrastive DPO pairs on top of personalized promptingQwen2.5-1.5B-Instruct · Multifaceted Benchwin rate: 39.29 -> 74.60Improved
train with MIPO on top of personalized promptingLlama-3.2-1B-Instruct · Community Alignmentwin rate: 78.00 -> 93.67Improved
train with MIPO on top of personalized prompting for the 7B modelQwen2.5-7B-Instruct · PRISMwin rate: 58.25 -> 58.76No clear effect
add MIPO contrastive pairs instead of SFT on the model's own chosen responsesLlama-3.2-1B-Instruct · ARC-Easyaccuracy: 34.13 -> 51.87Improved
train with MIPO without any verifierLlama-3.2-1B-Instruct · Reasoning/MCQ average (GSM, SVAMP, ARC-E, ARC-C)average accuracy: 37.72 -> 45.25Improved
train with MIPO on strongly math-tuned 7B modelQwen2.5-7B-Instruct · Reasoning/MCQ average (GSM, SVAMP, ARC-E, ARC-C)average accuracy: 91.71 -> 92.22No clear effect
MIPO versus RLVR with ground-truth rewardsQwen2.5-3B-Instruct · Reasoning/MCQ average (GSM, SVAMP, ARC-E, ARC-C)average accuracy: 81.45 -> 87.86Improved
SFT on the model's own outputs without contrast (ablation)Llama-3.2-1B-Instruct · Reasoning/MCQ average (GSM, SVAMP, ARC-E, ARC-C)average accuracy: 37.72 -> 35.48Hurt
replace reference-policy mutual information (DPO) with online MI reward via PPO (ablation)Qwen2.5-1.5B-Instruct · Community Alignmentwin rate: 78.83 -> 39.67Hurt
add RLAIF reward at 1B scaleLlama-3.2-1B-Instruct · PRISMwin rate: 72.17 -> 56.87Hurt
apply MIPO to improve output diversityQwen 1.5B · Personalization tasksSelf-BLEU-4: 0.310 -> 0.256Improved
use random or missing-context negatives instead of reshuffling chosen responses (ablation)MIPO · Personalized instruction-followingwin rate: random +1.3-13%, missing context +3-30% over reshufflingImproved

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