Using AI to write scripts

so now instead of one thing you don't know how to do, you have two things you don't have any clue about. sure, go on, i'll count the time to the first "oh, why this doesn't work" question.

If it gets the job done for your mini-hobby, sure. I mostly agree with @vyruss otherwise.

I have never actually used any LLMs myself but I'm not doomsday about 'em either.

Outside of Doom, some people use them as a glorified auto-complete integrated into their code editor, at least when they're typing out something obvious or repetitive; but I've also heard some people stop it after experimenting with it and turning it off, and they notice their brain lag because their muscle-memory has been trained to wait for a response.

Other people suggest they're good for auto-generating unit tests for your code that you wrote yourself. That's a good use. Probably doesn't apply to your Doom stuff though.

In answering questions, I have a love-hate for them. They boil things down to a median in a knowledge base, or just boil down the "safest" denominator. "Hey bot, will I ever find love?" Answer:
If I'm searching for an answer, I want a smorgasbord of varying hot takes.

Pertaining back to Doom: When it comes to creativity, my only gripe is that it ends up with a lot of "I made dis" situations. Even stuff like photoshop in the past had a lot of filters and tools (hard algorithms, not even trained models) that just "do stuff for you", and I have a petty side of me that feels like some artistic integrity is lost when using them. And, especially when leveraging machine learning, the line blurs for who/what to attribute the work to (how much of the art was actually you). There's something just more cool and badass about hand-etching the crap out of everything. Same in coding. Those Buddhist monks who painstakingly make sand Mandalas, only to smash them afterwards... *shivers*.

But yah, I wouldn't have shame in doing what you're doing. It sounds modest enough. It's low profile, and it's not like you're asking Copilot to build missile defense logic or anything.

Learning language models can interpret and create custom DECORATE? How much complexity can it handle?

A fair amount, you just need to provide enough context for it to "imitate" the syntax. Depending on the LLM you are using, its context window size may or may not be large enough to fit a nice example DECORATE in there.

In other words: If you can paste in an example, the LLM will do a much better job at generating new code for you based on your descriptions.

My speculation about LLMs is that there is a latent "meaning" space below the ordinary semantic space of language, and LLMs figure out how to map the two. And since they have an internal representation of the meaning space, they can create transforms between the two pretty easily. So once you give it the ability to transform "syntax" to "meaning", it can extend the concepts fairly easily.

It is a good way to help get to the root of a problem with coding, by copy pasting errors. I figured out a lot of CMD, Ubuntu, and powershell stuff I had to do, with it. It is really amazing.

I've created projects that are drawn pixel by pixel from scratch (like 95% of the patches from Aliens: Ultimate Doom, as well as sprites). But now I'm using it to help me make textures. You have to weed through a lot of generations and still do a fair amount of manual manipulation. But it is still an enjoyable endeavor and liberating from being bogged down reducing the amount of time it takes. Plus you can upload pictures too and ask for suggestions on what could make it look better to capture what you are intending.

Sometimes it creates bullshit which is very frustrating. But persistence is the key. It is like an idiot suvanet,  perhaps change up the prompt to more direct specifications.

I have my doubts about AI writing a script that's actually functional.

First screw ChetGPT. Don't give them your money. Their AI is trained on literal lies and refuses to give responses you ask for.

Here Kontra asked it for day-night cycle ACS code and it generated simplistic nonfunctional crap. I didn't even notice or remember this post with the code until after I programmed the huge amount of functional code in Friday the 13th Jason's Doom.

If anything, get X(formally Twitter)'s Grok, that's not chained down by communist woketards.

I agree there's nothing I could call true intelligence or creativity (at least not at this point). I just intend to say, I suspect meaning can be represented in a high-dimension vector space which allows mathematical transformations to modify meanings directly, and those modified meanings can be transformed back into language. Yes it is a "magic trick" in a sense.

The key point: There is no new information being generated in such a system, only transformations on existing concepts.

I probably will stop now to not go off the rails of the thread too much...