I don't think how much LLMs have disrupted other AI techniques is talked about enough.

Starting from an LLM and fine-tuning it to a specific domain can work significantly better than training old school ML models. As inference costs come down this will become even more commonplace.

I worked with several folks that worked in NLP and translation models for years at Google Search. LLMs completely wiped out their entire field since they can do it significantly better than their old techniques.

https://x.com/DiscussingFilm/status/1910832709477212302

Remember Hidden Markov Models? Pepperidge Farm remembers.