Reverse engineering GTA San Andreas with autonomous LLM agents [video]
agents autonomous
| Source: HN | Original article
A new video from YouTuber dryxio shows autonomous large‑language‑model (LLM) agents tackling the long‑standing “gta‑reversed” project, a community effort to recreate Rockstar’s 2004 classic Grand Theft Auto: San Andreas in clean C++. The agents, powered by OpenAI’s Codex and other LLMs, navigate the original binary, generate function signatures, and iteratively replace undocumented assembly with human‑readable code, all without direct human intervention. The demonstration, posted alongside a link to the project’s GitHub repository, marks the first time an AI‑driven pipeline has been applied to a full‑scale commercial game engine.
The significance extends beyond a nostalgic title. Reverse‑engineering legacy software has traditionally required teams of specialists painstakingly decoding obscure machine code. By delegating routine analysis and stub generation to LLMs, the process accelerates dramatically, opening the door to systematic preservation of aging games whose source code is lost or locked behind proprietary licences. For modders, an open‑source San Andreas engine could enable deeper gameplay tweaks, performance improvements, and ports to modern platforms. For the broader software‑engineering field, the experiment validates LLMs as viable assistants for “software archaeology,” a niche that includes security audits of legacy systems and migration of legacy code to maintainable languages.
The next steps will reveal whether the community can scale the approach to other Rockstar titles such as Vice City or GTA III, and whether the generated code can meet the performance and fidelity expectations of the original. Watch for updates from the gta‑reversed maintainers on code‑coverage milestones, for new videos documenting the agents’ learning curves, and for any legal response from Rockstar concerning the recreation of its engine. If the experiment proves robust, autonomous LLM agents could become a standard tool in the preservation and modernization of digital heritage.
Sources
Back to AIPULSEN