$T^2$ of Thoughts: Temperature Tree Elicits Reasoning in Large Language Models

TL;DR

This paper introduces Temperature Tree ($T^2$) prompting and a heuristic algorithm, $T^2$ of Thoughts ($T^2oT$), to enhance reasoning and decision-making in large language models by dynamically adjusting search parameters, improving accuracy and versatility.

Contribution

It presents a novel $T^2$ prompting method and heuristic algorithm that dynamically adjust search parameters, notably temperature, to improve reasoning in large language models.

Findings

Hybrid $T^2oT$ improves single-solution accuracy.

Adaptive $T^2oT$ enhances multi-solution generation.

Fixed search depth with $T^2oT$ offers reliable problem-solving.

Abstract

Large Language Models (LLMs) have emerged as powerful tools in artificial intelligence, especially in complex decision-making scenarios, but their static problem-solving strategies often limit their adaptability to dynamic environments. We explore the enhancement of reasoning capabilities in LLMs through Temperature Tree ($T^2$) prompting via a heuristic algorithm, termed as $T^2$ of Thoughts ($T^2oT$). The primary focus is on enhancing decision-making processes by dynamically adjusting search parameters, especially temperature, to improve accuracy without increasing computational demands. We empirically validate that our hybrid $T^2oT$ approach yields enhancements in, single-solution accuracy, multi-solution generation and text generation quality. Our findings suggest that while dynamic search depth adjustments based on temperature can yield mixed results, a fixed search depth, when coupled with adaptive capabilities of $T^2oT$, provides a more reliable and versatile problem-solving strategy. This work highlights the potential for future explorations in optimizing algorithmic interactions with foundational language models, particularly illustrated by our development for the Game of 24 and Creative Writing tasks.