LogFAQs > #970282570

[LFAQs-redacted-quote]

Yes, our civilization is definitely on course for a future in which the rich and well-connected will suddenly develop real empathy for other people and do an about face to make sure society functions well for most people in the face of rapid technological advancement.

Turtlebread posted...

ChatGPT is scarily helpful

Not at this point it isn't, at least not if you aren't getting lucky. A lot of the answers it gives you contains an eclectic mixture of obviously and subtly wrong details. Although the technology behind ChatGPT is impressive, it's really automating the process of a Google search and copying and pasting a random high up vote answer from Stack Overflow or something like Wikipedia for more general questions. It's ability to do things like write essays is more impressive, but all the examples I've seen that came straight out of it were the kind of thing that would barely get you a passing grade, if that, in college.

Sad_Face posted...

Use AI to come up with a solution to give men a purpose and goals to be productive to society.

Assuming AI can continue advancing at the rate people expect (we don't, after all, know the ultimate limits of technology), and assuming that civilization stabilizes and becomes equitable instead of dystopian, I have some trouble imagining what purpose and goals might be invented for human beings when AI seems to be on course to able to do everything we do, only faster and better. So far, the bulk of people don't seem to care who made what or how.

shironinja posted...

this technology is like having a child who you can bounce things off of as it is continuing to learn and become something new.

That's not exactly true. I'm not sure of the specifics of ChatGPT, but other things like Stable Diffusion aren't actively learning anything. The incredibly wide range of output you see is being generated by different models that have been trained using the technology. The base model was trained on 5 billion images, and then you can train more specific models to generate more specific things (like Waifu Diffusion, which is a model for generating anime style images, or there is one for Simpsons characters for example).

Sad_Face posted...

Bro, get onboard and start mastering the available tools now.

This is an interesting statement. What "mastering" do you envision here? There has been much talk of "prompt crafting" by the Stable Diffusion community as a means of deflecting criticism that no skill or hard work is involved in creating AI Art, and as a means of saying that AI artists really are creating their piece.

I've used Stable Diffusion quite a bit on my own system. Prompts typically consist of three parts: a sentence or series of phrases that vaguely or specifically describes the subject, like "an astronaut riding a horse"; a word salad of things like "highest quality", "masterpiece", and "trending on art station", the goal of which is to single to Stable Diffusion to make something better or to adjust the style slightly (by providing an artist's name, for example); and a negative prompt which tries to filter out stuff nobody wants to see like random floating text.

The only other dials and knobs consist of the guidance slider (one way for more literal, another for a less literal reading of your prompt) and the number of steps. Both the negative prompt and the quality word salad are things people are copying and pasting from Reddit posts and blogs. That just leaves the words describing the subject. You could do a fun mind map to brainstorm something...or you could just go to one of the many galleries that already exists online and copy a prompt from a piece that is close to what you want. Either way, you iterate in mere minutes until you find settings that produce images sort of like you want, and then you crank up the number of generated images until you have pages of finished product to look at and choose one you like.

Even the people behind Stable Diffusion don't think prompt engineering will ever be a job, because as the technology improves, it will only become easier to make the software understand what you want. The process of making AI art is nothing like other forms of digital creation, like say making materials in Substance Designer. The computer does everything for you, and what it isn't doing is simple with very few options or decisions to make. Try it yourself if you think I'm being a luddite. There is even a neat PDF guide that explains how this all works with a low page count, large font, and big pictures. You don't even need to install Stable Diffusion yourself---just go try out one of the versions available online. Your output won't look good without extras at the end of your prompt or a good negative prompt, but those are a literal Google search away.

People who think AI "tools" are going to create more jobs like previous rounds of automation really need to try it for themselves and see how it actually works.