the whole infrastructure feudalism thing for one. i don’t think AI discourse will ever get even remotely productive unless it is fully self-hosted and self-trained local stuff (kinda like arduino and raspberry scenes in a way) and given the hardware prices and its sheer inadequacy for scalable AI computations - that’s not going to happen any time soon.

the blackbox nature of AI models is close second. we have next to no idea what kind of data was used to train a model, how it was collected, whether its reasoning is clear-cut, whether it is accuracy assessments are fine-tuned - it is a russian roulette and in many cases it leads to all sorts of hallucinations and inaccuracies. it is one thing when you get some wild takes for shits and giggles and it is another when you’re paying top tier buck and get some asshat mumblebawl.

LLM dominion is another thing to bitch about especially when it comes to generative AI. there is a lot of potential with image analysis and pattern dimensional combinatorics and the whole visual culture mining and recontextualization but writing prompts is not the way to deal with it. not just because it is knee deep in the plagiarism and exploitation but also that’s not how images works and it is not like company are willing to explore different venues of generative AI that go beyond LLM. generative AI in general is a missed opportunity of cosmic proportions. it could’ve been this culture machine thing that explores various cultural footprints and patterns but instead it just does stuff like that other stuff and sometimes it is funny how bad it is.

the nonexistent business model is a problem too. i work in consulting and ever since ChatGPT became a thing i get a good dozen of requests to research business opportunities for AI this or AI that. Can’t complain because it pays well every now and then but AI products are such bullshit from a business standpoint - you just can’t get away from it being a gimmick with little to no value outside of novelty factor.

my favorite case of late was with the company that wanted to REVOLUTIONIZE THE LITERARY TRANSLATION. the thing with literary translation is that it takes more than just translating the text head on - there’s a lot of decision making and busywork outside of text that would require a dozen or so different reasoning models working in tandem and this whole thing is the opposite of cost-effective from any standpoint. to the point it would be cheaper to just pay for actual translator’s education and given the potential budgetary requirements for this project - it would’ve been enough to pay for an education of couple of thousands of translators. but somehow these motherfuckers got first round even though their project is literally unfeasible.