"Using Chat GPT" Topic

Now I was at Hmmeerhead UK show Satutuday chatting to a guy who had made a figure using this system, and woundered if anybody could shed more light on the technique or had tried it?

He had gotten a picture of a soldire in the style he wanted, in his case a splele tunic. He gave the picture to Chat GPT asking it to creatate a picture in the same style but with diffrent weapons in its hand and a diffrent stance. Having got what he wanted he put the picyure into "meshey" It looks like this is it.
link

This gives an STL files to scale. Apparently Chat GPT can make the picture a bit chunkier before meshing to make it more robust at small scales.

Hopeless for historical figures as AI has no actual intellegence (I understan why it fails now so miserablely when I use it, seems tascks requireing act
alk intellegence are outside the scope of LLM's).

However when approximations are all that is required it may be of use.

The big problem with both image generation and STL/Mesh generation is that they are trained on thousands (proabably millions) of images and STLs scrubbed from the web, the creators of which were not asked for permission in the vast majority of cases. Which to me seems like thinking its ok to steal a penny from everyone's bank account- its wrong but it doesn't hurt each individual that much, so I can do it. But in aggregate it would ammount to millions. We'd still consider that punishable.

clibinarium, I've heard the argument before. But how do human artists--or actors, writers or composers--learn? By watching great actors, listening to great music and studying great paintings, books and screenplays, last I heard.

When's the last time some aspiring actor sent a check to the Alec Guiness estate? Does anyone punish SF writers for learning from early Heinlein? (And as author of The Rolling Stones, did Heinlein kick back anything to the author of "Pigs is Pigs"? Certainly David Gerrold didn't pay him for "The Trouble with Tribbles," and that was close enough to worry the network.)

Let's not hold the machines to a legal/ethical standard we ourselves couldn't meet.

I use Chat GPT at work sometimes – kind of like having a simple minded assistant – useful but you gotta check their work out

I don't buy the stealing argument in general, though in some particular cases there has been some wrongdoing.

But to the larger point, I've watched a lot of videos about AI generated STL files. I have experimented a little. I'm dropping it for now. In a year, give or take six months, there will be a fun new aspect of our hobby, especially for those of us less artistically gifted.

For the moment you have to bounce around between different services (text to image, image to image, image to STL, STL to usable STL), some of which are crummy, and some of which cost money so I haven't tried them but I'm suspicious. You're still going to put in a lot of time just developing the workflow, finding the services and techniques you like.

This is going to work great in the future, but I'm going to let other people blaze the trail and then follow later.

Meanwhile, I'm having lots of fun with TinkerCAD and Blender.

When I was in a pinch I did use, I think it was Gemini, to make some paper miniatures. I needed some specific goblin pirates. I described each of them, asked for front and back views, did them several times, took the best, printed them out, folded them over, boom – I'm ready for the RPG session. This let me get figures with the exact equipment and expression for the NPCs.

Someday we'll be able to do that for resin minis, but right now it would require so many hours and so many subscriptions that I'm not interested, I have too many other cool things to do.

Robert, they learned FROM other artists, they don't slavishly copy what they did and claim it as 'original'.

Gildas Facit +1

Frederick +1

robert piepenbrink said:

clibinarium, I've heard the argument before. But how do human artists--or actors, writers or composers--learn? By watching great actors, listening to great music and studying great paintings, books and screenplays, last I heard.
[…]
Let's not hold the machines to a legal/ethical standard we ourselves couldn't meet.

But an LLM isn't a human learning a trade, and it doesn't have rights. It's an object owned by another human, operating in an economic system which has evolved over thousands of years to balance individual human rights with human welfare, and it completely upends the system that helped create it. Nobody can force you to give up your knowledge and enable your own displacement and impoverishment, but that is exactly how LLMs have been created – unwittingly, by the input of millions of individuals, without their knowledge or permission.

If somebody created a machine that could create piles of gold by sucking all the oxygen out of the air, would we let him use it? After all, nobody "owns" the air, and the gold-maker spent a fortune and a lifetime making the machine, what right do we have to stop him? But of course we would stop him; such a machine would be exploiting a gap in the rules of the economic system that helped create it but doesn't yet have a way to control it, benefit from it, or prevent its own destruction from it. LLMs are operating in a similar gap. Humans must be able to control them, or the majority of us will soon be controlled by them.

- Ix

Frederick said:

I use Chat GPT at work sometimes – kind of like having a simple minded assistant – useful but you gotta check their work out

It is still easy to get flawed answers and useless output from LLMs, and you do have to double-check nearly everything, but don't let that make you complacent. They are evolving faster than anything ever seen in history. LLMs have already grown in capacity by leaps and bounds in the last 6 months, and the pace of change is accelerating. Try again in 4 months, then 3 months, then 2 months, then every month after that. With the right combination of 3D printer and AI interface you should be able to stop paying businesses run by sculptors and casters by the end of next year.

Most of last year I was unable to find any good uses for LLMs, but now I use them everyday for work and personal computer/network development. My non-engineer software-industry coworkers are already getting robot-made custom computer-mounting pieces without ever consulting the industrial design engineers the company pays to do that kind of work. My company has nice stuff that regular people can't afford, but publicly available tech of this sort can't be too far off.

Some tips:

Pay for a subscription and get access to a more advanced model. You get better results. What you get for free is the slower, older, dumber versions of an AI. "First one's free" and all that.

Ask AI how to use AI. You can learn a lot from AI about itself.

Ask each AI about other AIs. I am finding it useful to ask each LLM about all the other LLMs I try to use, and compare/contrast the different answers – kind of like piecing together an event from the reports of multiple witnesses. It's also been useful to ask each LLM to recommend other LLMs for the kinds of tasks I can't get done on this LLM (whichever one I'm talking to).

Try as many different LLMs as you can find out about. Each one has strengths and weaknesses. I use ChatGPT for a lot of general questions and research, Claude for a lot of my technical work, Venice for questions that other AIs stumble over, Gemini for nothing because it's slow and useless, Grok for nothing because it hallucinates constantly, and I stopped trying to get Midjourney to make an image I want to look at (still on the hunt for a better image maker). And I feel like I've only barely gotten started.