I'VE PUT A COMPLEX CODEBASE INTO A SINGLE 120K-TOKEN PROMPT, AND ASKED 7 QUESTIONS GPT-4 AND GEMINI 1.5. HERE ARE THE RESULTS!

Summary from the TLDR Web Dev newsletter

This article compared LLMs Gemini 1.5 and GPT-4-Turbo-128K on their ability to understand a complex 120K-token codebase. Gemini 1.5 outperformed GPT-4 in almost all aspects, demonstrating a better understanding of the code syntax, memory layout, bug fixes, and removed functions. While neither model achieves true AGI (Artificial General Intelligence), Gemini's ability to navigate large contexts and connect information is impressive.