This paper introduces knowledge-centric self-improvement for AI systems, where agents remain generic and disposable while a curated knowledge base is used for future tasks, leading to more inspectable…

This paper surveys 1,250 arXiv papers on self-improving AI systems, categorizing them based on what they improve and the degree of loop closure. It identifies a distinctive feature of self-evaluation…

GRASP introduces a gated, regression-aware framework for improving LLM agents by ensuring that every proposed skill edit improves performance on a balanced probe without degrading previously learned c…

The paper introduces BenchTrace, a novel benchmark designed to rigorously evaluate the self-evolution and reflection capabilities of LLM agents, revealing that current models struggle with accurate fa…

The paper identifies a linear predictive law linking the initial performance gap in on-policy self-distillation (OPSD) to the final performance improvement, allowing researchers to anticipate and tune…

This paper introduces Skill Self-Play (Skill-SP), a co-evolutionary framework for LLM training that bridges the gap between structured verification and open-ended exploration.

The paper proposes Self-Trained Verification (STV), a novel method that trains verifiers to catch self-generated errors by leveraging reference solutions, significantly boosting performance in both te…

The paper introduces SAVE, a framework that uses on-policy feedback and the value function to self-supervise and improve reward models, significantly enhancing RLHF performance across multiple benchma…

The paper introduces EvoNote, a self-evolving agentic framework that significantly improves the generation of evidence-grounded health community notes by utilizing an accumulated memory of past misinf…

This paper conducted a randomized controlled trial on Reddit to test the effectiveness of various deescalation strategies in reducing personal insults using automated replies.

The paper proposes a coherent inverse reinforcement learning (IRL) method to improve large behavior models for robotic control, achieving superior sample efficiency and performance on complex sparse m…

This paper introduces Capability Self-Assessment (CSA), a crucial ability for LLMs to recognize their limitations, and demonstrates that reinforcement learning is an effective method for teaching this…

SIRI introduces a self-internalizing reinforcement learning framework that allows LLM agents to autonomously discover and integrate reusable skills directly into their core policy, significantly impro…

This paper proposes two strategies to improve feedback efficiency of reinforcement learning from human feedback (RLHF) in diffusion models.

SkillSmith is a synergy-aware framework that jointly co-evolves skills and tools, significantly improving self-improving agent systems by modeling skill-tool interactions and diagnosing failures.

This paper introduces RSIBench-Data, a controlled benchmark for evaluating data-centric research capabilities of LLM agents.

ReSkill is an RL-in-the-loop framework that reconciles skill creation and policy optimization by automatically creating, testing, and refining modular skills alongside the agent's policy learning, lea…

SkillC introduces a Contrastive Skill Credit Assignment (CSCA) framework to enable LLM agents to autonomously internalize skills during training, significantly outperforming existing methods without r…

SCOPE introduces a data-free self-play framework that co-evolves a task-generating Challenger and a document-answering Solver, significantly improving open-ended performance on language models without…