I think this is where a lot of the divide in AI opinions comes from - focusing on many ultraspecific incremental improvements for X feature of Y model, without really understanding why we're driving so hard at improving the tech at a high level besides the fact that the math is interesting and the outputs stimulate our minds. What direction will the greater world move in as a result of being able to transfer the style of van Gogh onto a photo of a corgi in a few seconds? I don't think it's a question asked enough and the impact will be underestimated, more so than the words "a fluffy corgi in the style of van Gogh" let on. As you say I think the outcomes will be unexpected to some of the people involved in making it happen. And I don't think new AI research is incentivized to include humanities sections about the higher-order effects of their formula-heavy papers about optimizations and sampling methods.

Technological progress is kind of like a boiling frog scenario to me, where the temperature was ratcheted up with the proliferation of LLMs and diffusion models for many more people to take notice. And I don't want to use the two-letter acronym when speaking of technological progress specifically, since at this point it's too loaded a term. (I think some use it as a label for "tech that's progressed too far".)

But I also think this implies that past some threshold it leads to negative outcomes for some actions to be made too easy to accomplish with technological progress. The limiting factor of being stuck in a state of inefficiency prevented them until the moment those barriers were dissolved. Some of those barriers are implicit and invisible and I think will only be appreciated in hindsight when we become an advanced enough society to finally remove them and see what happens.

When the cost of copying public artists' styles was driven to near zero, suddenly thousands more people were taking part in what is deemed by some of those artists to be plagiarism. Only this time it's marketed with terms like "style transfer" and "artist prompt studies" and "democratization" and "hey, now we can create N new images M% faster with these matrix multiplications and these miscellaneous optimizations, and our paper will now discuss these in technical detail..."