Coding Triangle: How Does Large Language Model Understand Code?

TL;DR

This paper introduces the Code Triangle framework to systematically evaluate large language models' understanding of code across analysis, implementation, and testing, revealing their strengths and limitations in coding tasks.

Contribution

The study proposes the Code Triangle framework for comprehensive evaluation of LLMs in coding, highlighting the importance of diverse data and model mixtures for improvement.

Findings

LLMs can form a self-consistent system across analysis, implementation, and testing.

Model errors tend to cluster due to training data biases.

Incorporating human data and model mixtures improves performance and robustness.

Abstract

Large language models (LLMs) have achieved remarkable progress in code generation, yet their true programming competence remains underexplored. We introduce the Code Triangle framework, which systematically evaluates LLMs across three fundamental dimensions: editorial analysis, code implementation, and test case generation. Through extensive experiments on competitive programming benchmarks, we reveal that while LLMs can form a self-consistent system across these dimensions, their solutions often lack the diversity and robustness of human programmers. We identify a significant distribution shift between model cognition and human expertise, with model errors tending to cluster due to training data biases and limited reasoning transfer. Our study demonstrates that incorporating human-generated editorials, solutions, and diverse test cases, as well as leveraging model mixtures, can substantially enhance both the performance and robustness of LLMs. Furthermore, we reveal both the consistency and inconsistency in the cognition of LLMs that may facilitate self-reflection and self-improvement, providing a potential direction for developing more powerful coding models.