PDFs Guides?

[Kontra Kommando]

Now that ChatGPT 4.0 has a tool that allows users to upload and examine PDFs, I think it would be great if there was a resource that provides tutorials in PDF format. The would allow users to look up what they need in short order, rather than sifting through a wiki. I find myself making my own PDF guides by Print-saving the webpages into PDF format, and combining it.

Just a suggestion, and I think it would be helpful.

If there already is a link to PDF guides, it would be greatly appreciated if someone could point me in the right direction. Thanks!

[Darkcrafter]

Oh yeah, having an ACS-GPT or ZScript-GPT right in Doom Builder would have been awesome, perhaps?

[Kontra Kommando]

Might as well, I think that would be great, even MS paint is getting integrated with AI.

[Kontra Kommando]

Speaking of Doombuilder. I have a PDF guide for 1.2

Does anyone have one for Doom Builder 2?

[ramon.dexter]

No matter what you feed, AI will never be a better mapper than human...

[Kontra Kommando]

That's certain the case for now, but what I mean is a pdf guide I can upload so I can quickly reference what I need and talk to in plain English so I can build up maps myself.

90% I've learned about modding/editing Doom comes from tinkering, really.

[SanyaWaffles]

I don't like the idea of feeding stuff to AI, even if it's just reference materials.

[Kontra Kommando]

I know there is apprehension to AI. But, I remember when people used to be very apprehensive over using the internet for stuff like shopping, and sharing credit card information. There was also a huge taboo over meeting people online for relationships. Today, both of those things are more common than doing them in the real world. Eventually the same will be true for AI, and the many applications it will have in society.

[Jekyll Grim Payne]

Doom Builder 2 is ancient. It's several generations behind the current map editor, which is Ultimate Doom Builder. UDB comes with built-in guides accessible with F1. They can be printed from the program, so you can use a virtual printer to conver them to PDF.

Attempts at using ChatGPT for coding something in GZDoom have already been made, many times, and they can't produce functional code. That would require a lot of specific training (which hasn't been done) and a lot of specific data (of which there isn't nearly enough available). Neural networks are not magic, they can't do the work for you unless you teach them first. As it is now, there are plenty of good applications for neural networks, and game design isn't one of them.

Semi-random map generation can already be very well handled by OBLIGE.

[Gez]

There are two things here that I really don't understand.

Now that ChatGPT 4.0 has a tool that allows users to upload and examine PDFs, I think it would be great if there was a resource that provides tutorials in PDF format.

What's the relationship with text generators? Do you want tutorials to be in PDF form so that they can be read by ChatGPT, instead of being read by people?

The would allow users to look up what they need in short order, rather than sifting through a wiki. I find myself making my own PDF guides by Print-saving the webpages into PDF format, and combining it.

I know everyone's workflow is different but I really don't understand how a PDF document can be better than a wiki. I guess with a PDF you can have everything in a single document instead of separate pages, so you can use Ctrl-F? But wikis have a search box and I did spend a lot of time creating redirects just to make the search feature more functional on the ZDoom wiki.

My experience of PDF vs. HTML is that PDF is always clunkier to use on a computer, as if it was a document format created for printing stuff instead of reading it on a screen... HTML being faster to load (plus I pretty much always have a browser open), and actually responsive to window resizing when wanting to have the documentation as a panel next to whichever application I'm working in.

[Kontra Kommando]

Indeed, I want the PDF version, so the GPT can search for things I need; multiple subjects at once, to combine it and give me feedback for more complex questions. Personally, I like it, because I can "talk" to the PDF, and have it explain certain aspects that may not be so apparent to me the way it is written. Sort of like breaking it down in layman's terms.

[Kontra Kommando]

Doom Builder 2 is ancient. It's several generations behind the current map editor, which is Ultimate Doom Builder. UDB comes with built-in guides accessible with F1. They can be printed from the program, so you can use a virtual printer to conver them to PDF.

Attempts at using ChatGPT for coding something in GZDoom have already been made, many times, and they can't produce functional code. That would require a lot of specific training (which hasn't been done) and a lot of specific data (of which there isn't nearly enough available). Neural networks are not magic, they can't do the work for you unless you teach them first. As it is now, there are plenty of good applications for neural networks, and game design isn't one of them.

Semi-random map generation can already be very well handled by OBLIGE.

Definitely needs a lot of trial and error with coding, but I've managed to use it professionally for stuff like SQL, and R script to success. Results improved when I uploaded PDFs with instructions, which is the reason I would like it for Doom-related content.

[inkoalawetrust]

I don't think there's that many PDF guides at all for Doom modding, so you'd have to convert existing guides to PDF yourself.

Might as well, I think that would be great, even MS paint is getting integrated with AI.

And that's pretty fucking stupid in my opinion, you use MS Paint to draw a shitty drawing or make super basic edits, linking it with an image generator is like putting a truck engine on a moped. If you want to use an image generator you can just go online and use one. Microsoft might as well turn MS Paint into a full image editor, or not since that would require them to actually try, unlike just taking the current opened image and putting it it through DALLE-3 (Which they didn't even make themselves). To be fair though I've heard that MS Paint finally has image layers though.


Also on the subject of training an LLM to act like a Doom modding guide and/or writing code, I'd suggest trying to use the models made by these guys instead, unlike GPT-4, they are actual open source and useful unlike those stupid plugins or whatever that OpenAI wants to push for ChatGPT. So you will probably have a much easier time fine tuning and training an LLM you can properly modify.

No matter what you feed, AI will never be a better mapper than human...

As far as I can tell he's not trying to train a neural net to make maps, but rather to write DECORATE, ZScript and ACS. No idea how training an AI to make maps would even work.