ทันเอไอ

Developer's AI Attempt to Recreate 'Thrust' (1986) Fails, But Uncovers Benefits in Legacy Code Analysis

An AI experiment to rewrite the classic game 'Thrust' was unsuccessful, but surprisingly became an excellent tool for 'software archaeology' to understand complex original code.

Software developer James Randall shared his experience experimenting with the latest AI, Claude, to recreate the 1986 classic game 'Thrust' for browser play. However, the initial results were disappointing. Randall stated that the AI produced unusable code, and he quickly concluded within the first 5 minutes that "AI can't build Thrust for me."

Instead of giving up, Randall shifted his approach from having the AI 'generate' to having it 'analyze'. He fed Claude the game's original disassembled source code (converting machine code back into human-readable language), prepared by Kieran Connell, asking it to explain the complex workings of the 6502 assembly language code (assembly language for the 6502 chip) used in computers of that era.

The AI proved excellent at this task. It meticulously explained and dissected various game components, from the physics system and ship controls to specialized programming techniques of that era, such as fixed-point arithmetic calculations and managing hardware limitations. Randall concluded that the AI's true value wasn't in recreating the game, but in serving as a tool to 'deep dive' and understand over 30-year-old original code.

The original game Thrust was created by Jeremy C. Smith for the BBC Micro computer and was recognized as one of the games with remarkable physics for its time. Sadly, Smith died in an accident in 1992. This experiment thus serves as a way to use modern technology to understand and honor the work of a programmer from the past.