Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Show HN: I made a GPT to generate retro NES Game cartridges

chat.openai.com

Post reply on HN
Print view View original

1–10 of 13 posts

Re: Show HN: I made a GPT to generate retro NES Game cartridges

#3

Post by ilaksh »

I actually assumed that this was actually creating ROMs.

I wonder if you could somehow use a large multimodal model trained on all of the existing NES software machine code along with manuals, forums and videos for each game. If it's not enough training data, maybe include other 6502 software, with the platforms encoded to differentiate?

Then you get a model that can take a description of a game and maybe a proposed screenshot and generate a new ROM. You train it further by testing the ROMs and giving feedback, starting with negative feedback for ROMs that don't boot.

Re: Show HN: I made a GPT to generate retro NES Game cartridges

#6

Post by BearOso »

ilaksh wrote: post #3

I actually assumed that this was actually creating ROMs. I wonder if you could somehow use a large multimodal model trained on all of the existing NES software machine code along with manuals, forums and videos for each game. If it's not enough training data, maybe include other 6502 software, with the platforms encoded to differentiate? Then you get a model that can take a description of a game and maybe a proposed…

The knowledge how to program on old systems like the NES is located sparsely on the Internet and there aren't many coding examples. A LLM would have a very difficult time producing a cartridge ROM with no single-byte errors. Even multiple iterations wouldn't iron out the errors. The closest thing you could do would be construct a template and have the LLM fill in the graphics or text, but that's not versatile at all.

Re: Show HN: I made a GPT to generate retro NES Game cartridges

#7

Post by pyinstallwoes »

BearOso wrote: post #6
ilaksh wrote: post #3

I actually assumed that this was actually creating ROMs. I wonder if you could somehow use a large multimodal model trained on all of the existing NES software machine code along with manuals, forums and videos for each game. If it's not enough training data, maybe include other 6502 software, with the platforms encoded to differentiate? Then you get a model that can take a description of a game and maybe a proposed…

The knowledge how to program on old systems like the NES is located sparsely on the Internet and there aren't many coding examples. A LLM would have a very difficult time producing a cartridge ROM with no single-byte errors. Even multiple iterations wouldn't iron out the errors. The closest thing you could do would be construct a template and have the LLM fill in the graphics or text, but that's not versatile at all.

Seems possible with LORA versions of them like stable diffusion uses it for modular recombination.

Re: Show HN: I made a GPT to generate retro NES Game cartridges

#9

Post by jerpint »

SCM-Enthusiast wrote: post #8

When i selected game of thrones, it generated an image hapily. When i asked it to do it again but different if gave me a copyright error. When i explained that it had already done it for me, it said, "Oh, Sorry! Here's what you asked for!"

I asked for South Park and it complained about copyright so I asked for east park where any connection to South Park would be completely accidental and happily obliged to create ~south~ east park
Post reply on HN