Reproduce “Large Language Models Are Neurosymbolic Reasoners” 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
Meng Fang, Shilong Deng, Yudi Zhang, Zijing Shi, Ling Chen, Mykola Pechenizkiy, Jun Wang
Researchers at the University of Liverpool and collaborating institutions developed an LLM-based agent that acts as a neurosymbolic reasoner, utilizing external symbolic modules to achieve an 88% average success rate in complex text-based games. This approach combines LLMs' natural language understanding with precise symbolic tools, addressing tasks like arithmetic and navigation in a zero-shot manner without requiring extensive training data.
Key insights
- LLMs can function as effective 'tool users' or 'module orchestrators,' leveraging their pre-trained knowledge for high-level natural language understanding and general reasoning while outsourcing precise, rule-based symbolic computations.
- Explicitly defining task-specific constraints within prompts significantly improves the LLM's ability to follow logical sequences and adhere to game mechanics, enhancing overall performance and efficiency.
- The neurosymbolic approach enables LLM-powered agents to tackle complex, symbolically-rich interactive environments effectively without the need for extensive, task-specific training or fine-tuning.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.