this post was submitted on 14 Mar 2024
33 points (92.3% liked)
Emulation - Retro Gaming In Style
986 readers
2 users here now
A community for discussing emulation and preservation of retro games. This community is intended for discussing the art of emulation, the tooling involved and retro gaming in general; it is not intended as a dump of ROM files.
founded 1 year ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
I could see AI being very helpful in the painstaking original target console RE'ing effort. A capable system could make short work in a process that takes years manually
But AI has no actual intelligence of it's own. It's not going to magically just figure things out. All it can do is spit back what it has been fed.
That's 100% not how AIs work. Not even LLMs.
The whole point of AIs is they work beyond their training data. Otherwise they couldn't do anything.
Those who say these things severely underestimate what AI is capable of or will be in short order or just don't understand how they work and why.
But setting that aside, I'm not saying we'll be able to feed AI a raw decompiled firmware and have it spit out a fully functional emulator in an hour.
But, in the near future we might be able to feed it raw decompiled firmware and it'll be able to map proprietary undocumented syscalls in a few minutes, that would be a big chunk of work that could take months of not years
A decent AI model could significantly lower the barrier to entry for emulator development from "A handful of elite hackers and programmers"
I see you don’t understand what an LLM is, how they operate or comprehend the kind and volume of training data that is required.
So you do not believe AI will eventually be able to near instantly code anything you desire including an emulator?
With current models? No. See my points above especially the one about the volume of data required.
Reverse engineering firmware is extremely niche, even more so for emulation. There are so few examples that current AI models wouldn’t have enough training data to work off of.