Post

Hello everyone. I don't use AI on my ASM, ROM hacking projects and my code is available on GitHub: github.com/VitorVilela7 All my projects are focused to be on the highest possible quality bar and you all know that my ASM code is heavily combined with tool code to make projects doable.

This is how I can even make SA-1 ROM hacks possible. TLR ("The Last Remapper") scans the entire ROM, aligns trace logs, usage maps and based on that it outputs remap ASM

#1 - Blatant comments everywhere. AI can't infer context just from code, they need a huge block of comments to be able to understand what is even trying to do. The so called "thinking". #2 - Non-sense opcodes. AI will try using other assembler syntax, mix Asar/WLA DX, Z80/65c816, etc.

#3 - Poorly optimized code. AI will try doing REP #$20 LDA $000000 CLC ADC #$0010 STA $000000 Vilela code is "REP #$21 LDA $00 ADC #$0010 STA $00"

I understand the world is in constant change and I've to use AI for tool coding, specially on the company I'm hired rn and the models constantly evolve (with so called "harness"). But without control, ROM hacks will transform from playable things to "AI TECH DEMOS" that are bloating TL everywhere