# Graph of Thoughts: Solving Elaborate Problems with Large Language Models

**Authors:** Maciej Besta, Nils Blach, Ales Kubicek, Robert Gerstenberger, Lukas Gianinazzi, Joanna Gajda, Tomasz Lehmann, Michal Podstawski, Hubert Niewiadomski, Piotr Nyczyk, Torsten Hoefler  
**arXiv:** [2308.09687](https://arxiv.org/abs/2308.09687)  
**Date:** 2023-08-18  
**Summary page:** https://www.summarizepaper.com/en/paper/2308.09687v1/

## Key points

- The paper introduces a framework called Graph of Thoughts (GoT) that enhances the prompting capabilities of large language models (LLMs).
- GoT allows the modeling of LLM-generated information as an arbitrary graph, going beyond existing paradigms like Chain-of-Thought or Tree of Thoughts (ToT).
- GoT represents units of information generated by an LLM as vertices in a graph and establishes dependencies between these vertices using edges.
- This approach enables the combination of different LLM thoughts to produce synergistic outcomes and allows for distilling the essence of entire networks of thoughts or enhancing thoughts using feedback loops.
- GoT offers several advantages over state-of-the-art methods on various tasks, including a 62% improvement in sorting quality compared to ToT and reducing costs by more than 31%.
- The extensibility of GoT with new thought transformations makes it suitable for spearheading new prompting schemes.
- The introduction of the Graph of Thoughts framework presents a significant advancement in enhancing the capabilities and performance of large language models.

## Summary

In their paper titled "Graph of Thoughts: Solving Elaborate Problems with Large Language Models," Maciej Besta, Nils Blach, Ales Kubicek, Robert Gerstenberger, Lukas Gianinazzi, Joanna Gajda, Tomasz Lehmann, Michal Podstawski, Hubert Niewiadomski, Piotr Nyczyk and Torsten Hoefler introduce a framework called Graph of Thoughts (GoT) that enhances the prompting capabilities of large language models (LLMs). GoT goes beyond existing paradigms like Chain-of-Thought or Tree of Thoughts (ToT) by allowing the modeling of LLM-generated information as an arbitrary graph. The key idea behind GoT is to represent units of information generated by an LLM as vertices in a graph and establish dependencies between these vertices using edges. This approach enables the combination of different LLM thoughts to produce synergistic outcomes and allows for distilling the essence of entire networks of thoughts or enhancing thoughts using feedback loops. By leveraging this framework, GoT offers several advantages over state-of-the-art methods on various tasks. For example, GoT demonstrates a 62% improvement in sorting quality compared to ToT while simultaneously reducing costs by more than 31%. The extensibility of GoT with new thought transformations makes it suitable for spearheading new prompting schemes. This work brings LLM reasoning closer to human thinking and brain mechanisms such as recurrence that are characterized by complex networks. Overall, the authors' introduction of the Graph of Thoughts framework presents a significant advancement in enhancing the capabilities and performance of large language models.