PAPER2LLM++: Continual Self-Evolution of LLMs from Research Papers

2026-10-02

AI Summary

Large language models struggle to autonomously assimilate recent research exposing their limitations, creating a disconnect between human discoveries and model evolution. This work proposes PAPER2LLM++, a framework that transforms academic literature into continuous supervisory signals, enabling self-evolution through evidence extraction, test verification, and safe updating. Notably, it introduces a novel "attempt-evaluate-commit" mechanism to ensure that integrating new findings neither induces catastrophic forgetting nor compromises general capabilities. Experimental results demonstrate that the model can progressively absorb new knowledge across consecutive failure sequences while retaining earlier gains. By effectively closing the loop from academic discovery to model evolution, this approach achieves robust and continuous improvement.

Abstract

Research on LLMs continually uncovers model limitations, their causes, and potential solutions. Yet these human discoveries remain largely disconnected from model evolution: an LLM does not automatically learn from new research about its own failures. We introduce PAPER2LLM++, a framework for continual self-evolution of LLMs from research papers. Rather than treating papers merely as knowledge to retrieve, PAPER2LLM++ uses the growing literature as a stream of evidence and supervision for model improvement. For each incoming paper, it extracts evidence-grounded findings, tests whether the reported limitation persists in the current model, and, when needed, converts the findings into candidate learning signals. A try-evaluate-commit procedure integrates an update only when it improves the targeted behavior without substantially forgetting prior improvements or degrading general capabilities. Across a sequential stream of research-discovered LLM failures, we show that models can progressively incorporate new findings while retaining earlier gains. PAPER2LLM++ thus takes a step toward closing the loop between human discovery and model evolution, enabling models to continually learn from research about their own limitations and improvements.

Problem

• Large Language Models
• Continual Self-Evolution
• Model Limitations
• Research Papers

Innovation

• Continual Self-Evolution
• Paper-to-LLM
• Try-Evaluate-Commit
• Catastrophic Forgetting Mitigation
• Large Language Models