Self-Evolving Deep Research via Joint Generation and Evaluation

Arxiv CS.CL (NLP/language models)

Proposes a novel framework where LLMs jointly generate and evaluate deep research reports, enabling self-evolution through iterative refinement. The method addresses the lack of explicit quality evaluation in current report generation by incorporating both generation and assessment within a single model.

llm

Arxiv CS.CL (NLP/language models)·Han ZhuChengkun CaiYuanfeng SongXing ChenSirui HanYike Guo

7 days ago