I have issues with these LLMs but what he is saying here isn’t wrong. If people can use photoshop to create art, this is fundamentally no different.
I think people mistake the llm image generators as capable of generating a final output. They are a playground and shouldn’t be consider the final product anymore than amateur developers using it to write an entire program. Peole that have tried to shipping the LLM output as a final product have face the level of ridicule it deserves.
The more I see these discussions and the way everyone seemingly talked past each other, the more clear it seems these LLMs will create a measurable divide between those who can use them to improve productivity and those who refuse.
Each side of the divide will point at the other side’s extremists as proof for why their side is the correct side. But as with most things, the path is somewhere down the middle.
You had me until the very end. I agree with most of what you wrote. But saying that each side points to the other side’s extremists smells like a very “centrist” take. What are the “extremist” views that the ultra pro-AI people are pointing to?
The massive issue is your first paragraph. The rest makes sense.
You should have found a different comparison. Photoshop, at least before they added Ai, was a tool that required competence and learning to use. You became better using and failing at it over time.
Llm as a bank of stolen work from great artists has its uses, like for initial concepts, which before I would do with Google images, looking at other people’s works. And like that, it just requires that I know what I’m looking for, not that I have any competence nor did it teach me anything other than learning what other people do.
The radicalism is bad but it’s inevitable since most people are trying to sell it as something that it is not.
Specially the employers. They don’t care if I’m getting better or learning anything, but that I’m outputting something faster and costing them less. It works for them, but it’s causing a great damage.
You seem to be ignoring some stuff specific to llms, but anyway I’m curious; which graphical llm is not based in other people’s work?
Perhaps stolen was a strong term, but they are all based on other people’s work right? When I used it I could clearly see the styles of some artists it was trained on.
I’m curious about that but even if it’s magically original to me that’s is still only useful to look at someone else’s work, even if is a computer work. It’s not a tool in itself I can master and improve my work.
Like fluid simulation for instance. You can light and render it well, but you can’t fully control it, because you’re not the one making it. When I need a very specific splash for a render I need to manually model it. And then llm is even worse because it controls the whole thing by itself.
I have issues with these LLMs but what he is saying here isn’t wrong. If people can use photoshop to create art, this is fundamentally no different.
I think people mistake the llm image generators as capable of generating a final output. They are a playground and shouldn’t be consider the final product anymore than amateur developers using it to write an entire program. Peole that have tried to shipping the LLM output as a final product have face the level of ridicule it deserves.
The more I see these discussions and the way everyone seemingly talked past each other, the more clear it seems these LLMs will create a measurable divide between those who can use them to improve productivity and those who refuse.
Each side of the divide will point at the other side’s extremists as proof for why their side is the correct side. But as with most things, the path is somewhere down the middle.
You had me until the very end. I agree with most of what you wrote. But saying that each side points to the other side’s extremists smells like a very “centrist” take. What are the “extremist” views that the ultra pro-AI people are pointing to?
I hope so! I am literally suggesting a centrist take. Don’t go all in but also don’t ignore its benefits.
There are a lot of people hyping it and a lot of people hating on it. There is value to them. Some over overstating it, others are understating it.
This does not mean that overstating delta is equal to the understating delta. The sentiment on LLMs isn’t a bell curve.
Refusing to use it to their own detriment.
Convincing others to not use it to their own detriment.
The massive issue is your first paragraph. The rest makes sense.
You should have found a different comparison. Photoshop, at least before they added Ai, was a tool that required competence and learning to use. You became better using and failing at it over time.
Llm as a bank of stolen work from great artists has its uses, like for initial concepts, which before I would do with Google images, looking at other people’s works. And like that, it just requires that I know what I’m looking for, not that I have any competence nor did it teach me anything other than learning what other people do.
The radicalism is bad but it’s inevitable since most people are trying to sell it as something that it is not.
Specially the employers. They don’t care if I’m getting better or learning anything, but that I’m outputting something faster and costing them less. It works for them, but it’s causing a great damage.
This is what I’m talking about. Not all LLMs are based on stolen works. Pick the worst and pretend that’s all of it.
But we all (most of us) know what they are selling doesn’t exist. We know they are full of shit.
There is no reason to bring it up for this discuss other than the glee from beating a dead horse.
This is true of any shit employer. LLMs isn’t the cause of this behavior.
You seem to be ignoring some stuff specific to llms, but anyway I’m curious; which graphical llm is not based in other people’s work?
Perhaps stolen was a strong term, but they are all based on other people’s work right? When I used it I could clearly see the styles of some artists it was trained on.
I’m curious about that but even if it’s magically original to me that’s is still only useful to look at someone else’s work, even if is a computer work. It’s not a tool in itself I can master and improve my work.
Like fluid simulation for instance. You can light and render it well, but you can’t fully control it, because you’re not the one making it. When I need a very specific splash for a render I need to manually model it. And then llm is even worse because it controls the whole thing by itself.