AI Paper Atlas

Research map / Large language models / Agents, code and math

Towards Execution-Grounded Automated AI Research

Chenglei Si, Zitong Yang, Yejin Choi, Emmanuel Candes, Diyi Yang, Tatsunori Hashimoto

ICML 2026 · Large language models › Agents, code and math · approach: LLMs for scientific discovery · 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
use execution-guided evolutionary search instead of the GRPO baseline recipeQwen2.5-Math-1.5B post-training · GRPO environmentvalidation accuracy (%): 48.0% -> 69.4%Improved
use execution-guided evolutionary search instead of the nanoGPT baseline124M GPT pre-training · nanoGPT environmenttime to 3.28 val loss (min): 35.9 min -> 19.7 minImproved
search for more epochs with Claude-4.5-Opusexecution-guided search · GRPO and nanoGPT environmentsbest performance per epochImproved
search for more epochs with Claude-4.5-Sonnet or GPT-5execution-guided search · GRPO and nanoGPT environmentsbest performance per epochNo clear effect
execute Claude-4.5-Sonnet ideas with GPT-5 instead of self-executionautomated executor · GRPO environmentexecution rate: 84% -> 42%Hurt
execute Claude-4.5-Sonnet ideas with GPT-5 instead of self-executionautomated executor · nanoGPT environmentexecution rate: 90% -> 78%Hurt
train ideator with GRPO on execution reward (average reward)Qwen3-30B-A3B · GRPO environmentaverage accuracy: 0.253 -> 0.343Improved
train ideator with GRPO on execution reward (average reward)Qwen3-30B-A3B · nanoGPT environmentaverage validation loss: 5.150 -> 4.066Improved
train ideator with GRPO on execution reward (max reward)Qwen3-30B-A3B · GRPO and nanoGPT environmentsmax reward per epochNo clear effect
RL training for 68 epochs (diversity)Qwen3-30B-A3B · nanoGPT environmentideas that are RMSNorm swap or EMA (of 128): 51 -> 119Hurt
add dynamic prompt with previous trajectories to RL (ablation)RL from execution reward · GRPO environmentrewardNo clear effect
add length reward to RL (ablation)RL from execution reward · GRPO environmentthinking length and total rewardMixed

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