@AdmSnackbar@mastodon.social @krystyna@toad.social @nixCraft@mastodon.social you may be right. One disruptive case for LLM that I see is a shift away from LLM to less hungry models. Speech to text model can be connected to intent recognition to summarisation, etc. Small, specialised models, also domain specific, for DNA, semantic video compression, etc. LLMs become enabler of these specialised models. Also LLMs could train LLM much cheaper. Basically cost efficiency gap could be closed.