Having said that, image generation to me was and still is magic. Not because I don’t understand it, but because I saw it as a way to get people with imagination but no skill to actually make art.
Having said that, the reality and how it is used is different.
I love it for asset creation and texture for games. As a programmer with 0 artistic skill it has been a god send to be able to do far higher quality UI while not bogging down my prototyping time.
But for like truely unique art... It's kind of a mess. Like try to get an AI to make a dwarf warrior with a Lance riding on the shoulders of an anthropomorphic cat person, who is dressed in monk robes.
AI struggles so hard with unique scenes like that... For now...
Somebody woke up and realized that LLM AI is just search rebranded to get more VC funding.
FOSS implementations need to be encouraged and supported at every turn on this, it's almost certainly the only way forward with any reasonably open ethical consideration.
It was a hype train for normies who wanted to become relevant in "tech" conversations without knowing shit. Like blockchain before it, and smartphone models before blockchain. And for other normies who dreamed that now that technology will finally make computers able to do what's told in natural language, like in the movies, and engineers obsolete.
For me it'll become more magical, when used appropriately.
I still use large-language models for fun. My fav phone reviewer is Marques Brownlee. I compared his best big phones in his phone awards to Claude 3 opus' best big phones – I asked Opus. I wanted to see the similarities and differences for fun.
I've been :) with the tight competition too. Claude 3 is making Gpt 4 and Gemini sweat.