This library can have a Python dev up and running in a few minutes
https://huggingface.co/docs/transformers/quicktour
It supports 13 applications of transformers and text generation is only one of them!
I have been working on text classification on and off for 18 years, for the most part systems better than ‘bag of words’ have been elusive. I was working at a startup maybe 5 years ago where the technology was not ready and today it is.
You see some blog post where somebody writes a prompt for ChatGPT that makes it do a task with 70% accuracy; often I show a simpler model (cheap to run on my own computer) 1000 examples (a day of work) and I am getting well above 90%. It is true that ChatGPT is a superhuman bullshitter, and it attracts the same simpleminded and indolent folks that were attracted to NFTs but there are a lot of people doing amazing things with transformers and not talking about it because they don’t want competition.
I don't think LLM progress/excitement/hype is going to slow down significantly. They're a breakthrough of the type that only comes along once a decade or two and leads to a phase change. What's happening in AI now resembles the social media landscape back in the early 2000s.