Step-by-Step Reasoning to Solve Grid Puzzles: Where do LLMs Falter?

TL;DR

This paper introduces GridPuzzle, a new dataset and evaluation framework for analyzing the reasoning chains of large language models in solving grid puzzles, revealing that current prompting methods do not improve reasoning performance.

Contribution

It presents a novel dataset, an error taxonomy, and a framework for detailed evaluation of LLM reasoning in grid puzzles, highlighting gaps in current prompting techniques.

Findings

Existing prompting methods do not enhance LLM performance on GridPuzzle.

Analysis reveals common reasoning errors made by LLMs.

The framework enables fine-grained assessment of LLM reasoning chains.

Abstract

Solving grid puzzles involves a significant amount of logical reasoning. Hence, it is a good domain to evaluate the reasoning capability of a model which can then guide us to improve the reasoning ability of models. However, most existing works evaluate only the final predicted answer of a puzzle, without delving into an in-depth analysis of the LLMs' reasoning chains (such as where they falter) or providing any finer metrics to evaluate them. Since LLMs may rely on simple heuristics or artifacts to predict the final answer, it is crucial to evaluate the generated reasoning chain beyond overall correctness measures, for accurately evaluating the reasoning abilities of LLMs. To this end, we first develop GridPuzzle, an evaluation dataset comprising 274 grid-based puzzles with different complexities. Second, we propose a new error taxonomy derived from manual analysis of reasoning chains from LLMs including GPT-4, Claude-3, Gemini, Mistral, and Llama-2. Then, we develop an LLM-based framework for large-scale subjective evaluation (i.e., identifying errors) and an objective metric, PuzzleEval, to evaluate the correctness of reasoning chains. Evaluating reasoning chains from LLMs leads to several interesting findings. We further show that existing prompting methods used for enhancing models' reasoning abilities do not improve performance on GridPuzzle. This highlights the importance of understanding fine-grained errors and presents a challenge for future research to enhance LLMs' puzzle-solving abilities by developing methods that address these errors. Data and source code are available at https://github.com/Mihir3009/GridPuzzle.