AI Paper Atlas

Research map / Large language models / Instruction and fine-tuning

On the Shelf Life of Finetuned LLM-Judges: Future Proofing, Backward Compatibility, and Question Generalization

Janvijay Singh, Austin Xu, Yilun Zhou, Yefan Zhou, Dilek Hakkani-Tür, Shafiq Joty

ICLR 2026 · Large language models › Instruction and fine-tuning · approach: Fine-tuning generalization and delta compression

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
evaluate weak-response-trained judges on newer, stronger generator responsesSFT, DPO, SFT+DPO judges (Llama-3.1-8B, Ministral-8B, Mistral-24B) · DeepScaleRconsistent accuracy (FutureProof)Hurt
retrain the judge on strong responses instead of weak ones (ablation)fine-tuned judges · DeepScaleR, strong-response evaluationconsistent accuracy (RefreshAdvantage): up to +7.6 points (DPO, Mistral-24B)Improved
use DPO rather than SFT when retraining on strong responses (ablation)fine-tuned judges · DeepScaleRRefreshAdvantage: SFT minimal gain; DPO up to 7.6 pointsImproved
add SFT loss on top of DPO (ablation)judge training · DeepScaleRRefreshAdvantageMixed
scale judge backbone from 8B to 24B (DPO) (ablation)retrained judges · DeepScaleRRefreshAdvantage: 4.3 -> 7.6Improved
evaluate strong-response-trained judges on older weak responsesDPO and SFT judges · DeepScaleRBackCompatibilityMixed
continually train weak-trained DPO judge on strong responses (ablation)DPO judges · DeepScaleRFutureProof / RefreshAdvantageImproved
continual training versus strong-only training, backward compatibility (ablation)DPO judges · DeepScaleRBackCompatibilityMixed
evaluate on questions unseen during training instead of seen questions with new responsesfine-tuned judges · DeepScaleRconsistent accuracy (QuestionGen)Hurt
use SFT instead of DPO for question generalization (ablation)fine-tuned judges · DeepScaleRQuestionGen dropImproved
move from math (DeepScaleR) to knowledge-heavy MMLU-Projudge shelf-life metrics · MMLU-ProFutureProof magnitudeImproved
retrain on strong responses with SFT+DPO, 24B vs 8B backbone (ablation)judges · MMLU-ProRefreshAdvantage: 3.2 -> 4.4Improved

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