Yuanhao Zeng, Shaocheng Lan, Ao Lu, Kuang Yan, Heng Lin, Xinang Xu, Yuan Luo, Yueting Wang (ShanghaiTech University) · Agent Evolution Study
The Resolution Hypothesis: Agent Evolution's Key Is Identifying Improvements
The paper argues that agent evolution, where a proposer writes candidate prompts, skills or memories and selection keeps those that score higher, often stalls because noisy evaluations cannot reliably identify real improvements. Across seven evolution methods and six models, using over 10M LLM calls, the authors find selection at candidate level is no better than random at separating helpful from harmful artifacts. They also find typical margins of a few percentage points are smaller than run-to-run variance, so published method rankings often rest on luck.
Open the source ↗Similar Research & Safety releases
- Predicting End-to-End Fine-Tuning Time for Generative Language Models
Dharma-AI (Hugging Face blog) · Fine-tuning Time Predictor · Week 48, 2026
- WorldRuler: Interactive World Modeling under Configurable Rules
Mingliang Zhai, Chuanhao Li, Jiaming Tan, Zhen Li, Yuwei Wu, Yunde Jia, Zian Meng, Yuanyang Yin (Alaya Lab) · Rule-driven world model · Week 48, 2026
- PROLA: Principal-Orthogonal Low-rank Adaptation
Minseo Yoon, Minseong Bae, Sojin Lee, Hyunwoo J. Kim · Weather LoRA adaptation · Week 48, 2026
- From Repositories to Training Environments: Scaling Long-Horizon Coding Agents with Environment Evolution Loop
IQuest (Yang Tian et al.) · Coding agent environments · Week 48, 2026
- Weight-Space Programming Doesn't Need Transformers
Hyunjong Ok and Jaeho Lee, Pohang University of Science and Technology · Weight-space architecture · Week 48, 2026
- Multi-Branch Transformer
Gaotang Li, Lingjie Chen, Chunyuan Deng, Zhiwei Xu, Hongzheng Chen, Christopher De Sa, Jingrui He, Yingheng Wang · Branching Transformer · Week 48, 2026
Published by Quanta AI Labs, an AI agency in India. Browse every week in AI News Weekly.