Reproduce “Agentic Knowledgeable Self-awareness” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Shuofei Qiao, Zhisong Qiu, Baochang Ren, Xiaobin Wang, Xiangyuan Ru, Ningyu Zhang, Xiang Chen, Yong Jiang and 3 more
KnowSelf, a method for LLM agents, enables dynamic regulation of reflection and external knowledge based on situational awareness. It consistently outperforms existing baselines on ALFWorld and WebShop datasets while utilizing a significantly lower knowledge rate, achieving performance comparable to or surpassing larger models like GPT-4o.
Key insights
- Explicitly training agents for 'agentic knowledgeable self-awareness' leads to a more efficient and effective deployment of cognitive resources compared to indiscriminate knowledge injection.
- Self-reflection alone can significantly improve agent performance, suggesting that agents are often constrained by learned planning patterns rather than an inherent inability to solve tasks.
- The decision to invoke 'Knowledge' or 'Reflection' emerges as a late-stage cognitive process, appearing in the final layers of the agent's transformer model.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.