How not to solve Jane Street's ASIC puzzle. Kinda (www.babush.me)

A recent exploration into Jane Street's ASIC puzzle highlighted the challenges and potential of using AI-driven strategies for reverse engineering. An individual leveraged ChatGPT and Claude Code to craft a fuzzer, aiming to navigate the complexities of Application-Specific Integrated Circuits (ASICs) design, particularly through analyzing GDS files and the connectivity of circuit elements. The attempt, albeit made post-deadline, underscored the intriguing intersection of AI tools and circuit analysis, showing that while AI can assist in extracting and modeling circuit behavior, it doesn’t guarantee success due to the inherent complexities involved, such as handling transistor-level details effectively. This endeavor is significant for the AI and machine learning community as it illustrates both the promise and limitations of using large language models (LLMs) in technical problem-solving. The results indicated that while fuzzing can sometimes yield quick insights, it often falls short without deeper understanding and careful instrumentation of the system, emphasizing the need for a balanced approach that combines AI capabilities with human expertise. The journey also revealed that automation in testbench simulations could drastically reduce execution times, demonstrating the potential for AI tools to enhance efficiency in engineering tasks, albeit with the caveat that challenges like reachability remain pertinent in such intricate domains.