AI Paper Atlas

Research map / Large language models / AI and society

Training AI Co-Scientists Using Rubric Rewards

Shashwat Goel, RISHI HAZRA, Dulhan Jayalath, Timon Willi, Parag Jain, William F. Shen, Ilias Leontiadis, Francesco Barbieri, Yoram Bachrach, Jonas Geiping, Chenxi Whitehouse

ICML 2026 · Large language models › AI and society · approach: Benchmarking and reproducibility practice · 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 rubric-based GRPO and 4B self-reward (ablation)Qwen-3-4B · validation, Claude-4-Sonnetrubric items satisfied (%): 12.0 -> 21.7Improved
use SFT on reference solutions instead of RL (ablation)Qwen-3-4B · validation, Claude-4-Sonnetrubric items satisfied (%): 12.0 -> 3.4Hurt
disable the KL penalty (ablation)Rubric RL with 30B RM · validation, Claude-4-Sonnetrubric items satisfied (%): 23.3 -> 29.7Improved
use stronger 30B MoE grader instead of 4B grader (ablation)Rubric RL with KL penalty · validation, Claude-4-Sonnetrubric items satisfied (%): 21.7 -> 23.3Improved
add reference solution for the grader (ablation)Rubric RL (4B RM) · validation, Claude-4-Sonnetrubric items satisfied (%): 20.0 -> 21.7Improved
add goal-specific rubrics to grading (ablation)Rubric RL 30B RM · validation, Claude-4-Sonnetrubric items satisfied (%): 20.9 -> 29.7Improved
add general guidelines to grading (ablation)Rubric RL 30B RM · validation, Claude-4-Sonnetrubric items satisfied (%): 19.8 -> 29.7Improved
add data filtering (ablation)Rubric RL 30B RM · validation, Claude-4-Sonnetrubric items satisfied (%): 28.8 -> 29.7Improved
train with rubric RL and compare to initial model with human expertsQwen-3-30B-A3B-Instruct · ResearchPlanGen-ML human studyexpert overall score (1-10): 7.31 -> 7.89Improved
apply medical-finetuned model to ArXivQwen-3-30B · ResearchPlanGen-ArXivrelative improvement: 17.5%Improved
train on Llama-3.1-8B with the same recipeLlama-3.1-8B · ML research goalsrubric score: 4.9 -> 18.5Improved
train on Gemma-3-4B with the same recipeGemma-3-4B · ML research goalsrubric score: 12.0 -> 16.7Improved

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