Where is that coming from? The chances that some random ps2 games code symbols are in the training data are infinitesimal. It's much more likely that it can understand code and rewrite it. Basically what LLM have been capable of for years now.

Parent is supposing w/o any experience. LLMs can see in hex, bytecode and base64, rot13, etc. I use LLMs to decompile bytecode all the time.