PaperCritic: RL-trained AI reviewers beat frontier models at judging paper soundness
CarnegieMellonU · hf · 2026-11-17
A paper from KAIST AI, CMU LTI, and LG AI Research trains AI reviewers focused solely on assessing the soundness of papers.
- Motivation: naive imitation of human reviews is flawed since reviewing serves multiple purposes (soundness, interestingness, clarity), the latter two subjective; this work targets the most objective and important criterion
- Method: reinforcement learning with an objective that jointly maximizes precision and recall of soundness review points, rather than imitating human scores
- Results: outperforms frontier models and other training approaches, and transfers to other review domains
A pragmatic direction for easing review loads amid exploding AI conference submissions.
More from Research
- Latent feedback cuts Transformer training compute by up to 46% with 4 lines of code — IFM · 2026-11-17
- SJTU's Calpha folds proteins from a general-purpose LLM's internal residue states — SJTU · 2026-11-17
- LoopFill: training-free speculative decoding speeds up looped Transformers — UIUC-CS · 2026-11-17
- AllocFlow: end-to-end flow matching turns raw market data into portfolio allocations — Annan Wang · 2026-11-17
- SPILLWAY: deterministic defense stops all templated prompt injections on tool-using agents — saintmachines · 2026-11-17
- VoxMel TTS: tokenizer-free autoregressive speech synthesis via Haar-factorized mel — VoxDance-Lab · 2026-11-17