I’m not entirely against LLMs as a tool, but I especially despise the image-based LLMs. They are certainly neat for some fun things. I’ve used them a little bit here and there for a dumb profile picture or a “I’m kinda thinking about this…” Brainstorm, but even in those cases I noticed the capabilities of the LLM and its tendencies quite literally pidgeon hole my artistic vision and push me in other directions that felt less and less creative. (Sidenote: I feel the same way about coding LLM tools. The longer I use them at any given time, the less creative I feel and it has a noticeable impact on my interest in the code I’m writing. So I don’t really use them much. Also I consistently manage to point out coding LLM code in PR reviews because it’s always kinda funky)
I’ve avoided using AI art tools for a while now. I’ll consider some limited use if the cost, billionaire ownership, blatant theft of real IP without compensation, and environmental impact problems are solved. (No, an “open source” model doesn’t solve all of these problems, especially since nearly all open source models are not truly open source and are almost always benefiting from upstream theft)
You know what I do like about AI art? I like the older Google machine learning art experiments from the mid-2010s. They invoked a strange existential curiosity. But those weren’t done with LLM’s.
Outside of LLMs, I like that there are some newer tools for editing that can do a better “lasso” select, that can mix and match into brushes as an alternative to something more algorithmic, the audio plugin that uses a RNN to simplify or expand upon an audio technique. Things that are tools that can be chosen or avoided and have nothing to do with LLMs.
I honestly cannot wait for this bubble to burst and for these tools to return to a cost that they’d need to be for these companies to turn a profit. A higher cost would eliminate all this casual use that is making people worse at research, critical thinking, and creativity, as well as make the art tools less competitive to just paying artists, even for scumbags wanting to cut the artists out. And it’d incentivize non-LLM, non-insanely costly ML techniques again instead of the current “LLMs for everything” nonsense right now.
I’m not entirely against LLMs as a tool, but I especially despise the image-based LLMs. They are certainly neat for some fun things. I’ve used them a little bit here and there for a dumb profile picture or a “I’m kinda thinking about this…” Brainstorm, but even in those cases I noticed the capabilities of the LLM and its tendencies quite literally pidgeon hole my artistic vision and push me in other directions that felt less and less creative. (Sidenote: I feel the same way about coding LLM tools. The longer I use them at any given time, the less creative I feel and it has a noticeable impact on my interest in the code I’m writing. So I don’t really use them much. Also I consistently manage to point out coding LLM code in PR reviews because it’s always kinda funky)
I’ve avoided using AI art tools for a while now. I’ll consider some limited use if the cost, billionaire ownership, blatant theft of real IP without compensation, and environmental impact problems are solved. (No, an “open source” model doesn’t solve all of these problems, especially since nearly all open source models are not truly open source and are almost always benefiting from upstream theft)
You know what I do like about AI art? I like the older Google machine learning art experiments from the mid-2010s. They invoked a strange existential curiosity. But those weren’t done with LLM’s.
Outside of LLMs, I like that there are some newer tools for editing that can do a better “lasso” select, that can mix and match into brushes as an alternative to something more algorithmic, the audio plugin that uses a RNN to simplify or expand upon an audio technique. Things that are tools that can be chosen or avoided and have nothing to do with LLMs.
I honestly cannot wait for this bubble to burst and for these tools to return to a cost that they’d need to be for these companies to turn a profit. A higher cost would eliminate all this casual use that is making people worse at research, critical thinking, and creativity, as well as make the art tools less competitive to just paying artists, even for scumbags wanting to cut the artists out. And it’d incentivize non-LLM, non-insanely costly ML techniques again instead of the current “LLMs for everything” nonsense right now.
It’s pleasant to see such a long post so full of good takes. Nice work!