Ai and generating an outcome

[Clint_L]

I played around with chat for something I am writing, and it was worthless. So not replacing me yet.

I've used it for plenty of things. Worked for me. Doesn't mean it's replacing me, though. Just one more tool in the belt.

 

[dragoner]

I've used it for plenty of things. Worked for me. Doesn't mean it's replacing me, though. Just one more tool in the belt.

That was maybe what I was hoping, though what I am writing is esoteric enough, probably too much, for it.

 

[the Jester]

I am surprised by the high number of yes vote here (especially on the first item, totalizaing 40+% so far) because given the tone of the discussions, I'd have thought the local crowd was much more against AI on principles rather than because of the legal uncertainties about scraping the Internet.

The principle being stood on by many of us is that AI is 'learning from' or imitating art without the permission and buy-in of the artists, so option 1 is not mutually exclusive with the principle in question. In fact, the whole premise is that the AI is learning from/imitating only those things that they have principled access to.

 

[Eyes of Nine]

I dont think most people care about ethic's in their consumption.

I kind of think ethics are relative

Like, I think it's ethical for AI to use already created content for training if given permission (which I read as PAYING) by the human artists who created the content

So we both think two different - and perhaps at odds - positions are ethical. Who is right?

 

[Clint_L]

I am not entirely convinced by requiring AI to have to be trained on public domain works, to be honest. It depends on the nature of the AI. Because human intelligences are trained on plenty of work that isn't in the public domain. I'm a big music buff, and almost all of pop music is pretty derivative.

 

[MoonSong]

View attachment 344211

That kind of reminded me of Atlanta Nights.

 

[MoonSong]

I am not entirely convinced by requiring AI to have to be trained on public domain works, to be honest. It depends on the nature of the AI. Because human intelligences are trained on plenty of work that isn't in the public domain. I'm a big music buff, and almost all of pop music is pretty derivative.

Human intelligence is quite different from Machine Learning. Machines lack free-will, desires and experiences. They cannot create novel ideas nor even internalize existing ideas. Training models are just a pure data derivative, at most a glorified -though quite sophisticated- sweat-off-the-brow product.

I'm just getting into learning some music theory. And yes, there's a lot of derivative and formulaic stuff, and a lot of inspiration comes from current music that isn't public domain. But only a human can associate certain notes in a certain rhythm with an emotion they might not even be able to put a name to. (And the irony is that music-making AI's didn't have to be trained on anything but public domain stuff to do interesting things and be of use)

 

[Jfdlsjfd]

Human intelligence is quite different from Machine Learning. Machines lack free-will, desires and experiences. They cannot create novel ideas nor even internalize existing ideas. Training models are just a pure data derivative, at most a glorified -though quite sophisticated- sweat-off-the-brow product.

I'm just getting into learning some music theory. And yes, there's a lot of derivative and formulaic stuff, and a lot of inspiration comes from current music that isn't public domain. But only a human can associate certain notes in a certain rhythm with an emotion they might not even be able to put a name to. (And the irony is that music-making AI's didn't have to be trained on anything but public domain stuff to do interesting things and be of use)

Sure. It's a great tool. The AI won't imagine the exact details of the ritual one of the PC in my group cast. It was a human creation by the players. And the human is specifying how to draw the individual elements needed to represent the vision in his creative mind, none of which are beyond the ability to be described by words (or specific tools). The AI won't invent human-walrus hybrid, sure, but it can generate images mixing a walrus and a man, under the guidance of the human that will correct it and select the best result until it fits exactly its vision of the walrusman, despite none of them existing before.

Seems to be the same with music: left to its own device it will generate elevator music, but if you've a vision of the music you want to create, you can use the tool to help you doing that "mind2music" part.

 

[trappedslider]

so I was looking at civitai and saw this as part of one model's description "This new model was fine-tuned using a vast collection of public domain images"

 

[Clint_L]

Really interesting article on research into how Large Language Model AI like Chat-GPTs produce text, specifically demonstrating that they are not just stochastic parrots (i.e. they have some level of understanding, and it is growing as the LLMs themselves grow):

New Theory Suggests Chatbots Can Understand Text | Quanta Magazine

Far from being “stochastic parrots,” the biggest large language models seem to learn enough skills to understand the words they’re processing.

www.quantamagazine.org