Graph of Thought: Boosting Logical Reasoning in LLMs

Created in August 30, 2023

2023   ·   llm   reasoning   ·   paper-review

This paper proposes representing intermediate reasoning steps as a graph rather than a linear chain. While Chain-of-Thought (CoT) forces reasoning into a sequence, graph-structured representations allow:

  • Branching: exploring multiple reasoning paths simultaneously
  • Merging: combining conclusions from different sub-problems
  • Cycles: iterative refinement of intermediate conclusions

The graph-of-thought framework generalizes CoT and Tree-of-Thought and shows improvements on multi-step logical reasoning benchmarks that require non-linear deduction.



Enjoy Reading This Article?

Here are some more articles you might like to read next:

  • The Expressive Power of Transformers with Chain of Thought
  • Landscape of Thoughts — Visualizing Where LLM Reasoning Actually Goes
  • Magellan — Guided MCTS for Escaping the Gravity Wells of LLM Creativity
  • PriorZero — Injecting LLM Priors into MuZero-Style World Models at the MCTS Root
  • SuperThoughts — Reasoning Tokens in Superposition