[talldayo]

> AI however, went from something most people had never heard of to full-speed exploitation.

To be fair, the timeline goes a bit further back. Markov chains had their moment in group chats a decade ago, it's hard to find a communal IRC room without someone's Markov bot lurking around. Then came BERT and GPT-2, both of which had their innocent days of getting finetuned into somewhat-usable form. After GPT-3 was the "oh we can sell this" moment, and all the cutesy, semi-efficient research that was being done got sidelined so people could make their own ChatGPT-killer. That is the disturbing part to me; so many people (OpenAI included) abandoned efficiency for marginal gains. This seems to be the trend going forward, likely exacerbated with GPT-5; we're going to need bigger boats (servers).