Narrowing the Complexity Gap in the Evaluation of Large Language Models

TL;DR

This paper introduces GeneBench, an automated method to add real-world complexity to programming benchmarks, revealing that LLMs' performance drops significantly under more realistic conditions, thus providing a more accurate evaluation of their capabilities.

Contribution

GeneBench offers a novel, automated multi-objective optimization approach to enhance benchmark complexity, addressing data contamination and overfitting issues in LLM evaluation.

Findings

LLMs' performance drops by 14.9%-60.5% on complex benchmarks

Performance decline persists even with few-shot prompting or fine-tuning

GeneBench's results are consistent with real-world bug repair performance

Abstract

Evaluating Large Language Models (LLMs) with respect to real-world code complexity is essential. Otherwise, there is a risk of overestimating LLMs' programming abilities based on simplistic benchmarks, only to be disappointed when using them in real-world settings. Recently, researchers explored the construction of more realistic benchmarks by mining or augmenting open-source repositories. Such solutions are usually task-specific. Data quality control from real-world projects can also be time-consuming and error-prone. More importantly, evaluating LLMs on fixed benchmark problems is subject to data contamination and overfitting. We propose GeneBench, an automated technique to add real-world complexities to any programming benchmark. GeneBench leverages a multi-objective optimization to increase the complexity of programming problems while maintaining the readability of code similar to real-world programs. Transforming four widely-used programming benchmarks using GeneBench and evaluating 13 LLMs (including two reasoning LLMs) on them shows a notable performance drop across all programming tasks (14.9%-60.5%, avg=35.2%), demonstrating LLMs' struggle under real-world complexities. The struggle persists even when LLMs are few-shot prompted or fine-tuned with examples from different versions of GeneBench, demonstrating the challenging nature of the problems. Finally, we show that the performance of the studied LLMs in bug repair is similar under GeneBench and SWE-Bench. This, along with the consistent reproduction of performance drop of all studied LLMs across four tasks under different versions of GeneBench, makes the technique suitable to evaluate LLMs without costly construction of real-world benchmarks.