LLMs predict geometry theorems using graph priors without gradient descent

Researchers have developed a new method for predicting geometry theorems using large language models (LLMs) without requiring gradient-based optimization. Their approach, called Theorem Precedence Graphs, encodes temporal dependencies from historical solutions into directed graphs to guide the LLM's search and overcome a scalability issue known as Structural Drift. This technique allows LLMs to function as structured planners, achieving 89.29% accuracy on the FormalGeo7k benchmark, which rivals state-of-the-art supervised models. AI

IMPACT Explicit structural priors offer a promising direction for scaling LLM-based symbolic reasoning.

RANK_REASON Academic paper detailing a new method for LLM-based symbolic reasoning. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• FormalGeo7k
• LLMs
• Theorem Precedence Graphs
• Ting Zhang

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →