AI mega thread

[quote=“Gryzor, post:485, topic:743”]

* Comparing prompting an LLM to holding a paintbrush confuses performing a task with commissioning one.

[/quote]

If you’re fundamentally “vibe coding” then sure. If you ask AI to produce a Z80 routine that multiplies an 8.8 fixed point number in HL by A, are you really doing anything other than what a C compiler might do?

It’s really not any more - anything that does peephole optimisation or full program static analysis is essentially rewriting code in ways the programmer has no control over. Heck even writing assembly language on a modern CPU isn’t really any guarantee that things execute 1:1 with the specified instructions.

P.S. I’m not sure there is any way to multi quote on mobile. The copy quote option doesn’t seem to work properly.

Except this isn’t encryption by the fact there is no transformation no algorithm. There are a lot of morons who claim to be cryptography experts and say it’s a OTP or its a Book Cipher which is hilarious because… it’s impossible to be both. It’s neither but shares some properties of each.

Yes, you’re right on that; I was being a bit simplistic, because I don’t think we need to descend into technicalities that much - but I’ll give you the point indeed. That said, asking an LLM for a textbook 8.8 fixed-point HL * A routine, reading every line, and dropping it into your engine is completely different from prompting a whole game into existence without knowing how it works. Used at that scale, an LLM is just an interactive helper.

The line for me is semantic guarantees and verification. A compiler’s optimizer is strictly bound - it can rearrange instructions, but it can’t change the mathematical result, and the same input always yields the exact same binary (unless I’m seriously misunderstanding something). An LLM predicts plausible tokens. It might succeed in that routine on one prompt and fuck it up on the next. That workflow only holds together if you already know enough Z80 assembly to audit the output yourself (which brings us right back to human authorship.)

You don’t copy, you select your text and then select “Quote” from the pop-up menu.

Thank you, not interested in the least in argumentation with someone calling a friend of mine a moron without having the slightest idea about them. Please watch your language next time. The only reason the comment was left unedited was for me to make this point here.

2 Likes

I didn’t call your friend a moron. I called those who claim it to be what it isn’t a moron.