Bringing a Warhammer to a Knife Fight (reorchestrate.com)

A recent exploration into the reverse engineering of an old Multi-user Dungeon (MUD) has leveraged modern Large Language Models (LLMs) to automate the process of decompiling lost code. The approach, which can also be utilized in tasks like converting legacy COBOL to contemporary languages, promises significant advancements in how we handle outdated software. By applying their capabilities in summarizing vast data sets and performing 'neural' translations between programming languages, LLMs, particularly Google's Gemini models, are proving effective in identifying patterns and renaming variables in decompiled code, facilitating the reconstruction of original logic. This project not only reveals the strong potential of LLMs in software engineering but also highlights how tools like the Ghidra decompiler and Gemini’s advanced context usage can enable more efficient and accurate code translation and comprehension. As the decompilation process iteratively improves with LLM assistance, previously forgotten functions can be reinstated into a human-readable format, making the implications profound for preserving and modernizing legacy systems. The melding of AI with software engineering could innovate how we approach both active development and the maintenance of older programs, ultimately reshaping the landscape of code sustainability.