There’s a third option in my humble opinion: it’s a bubble in some areas and not in others.
It’s a bubble for the amount of investment, which has no real justification, in general AI models.
Smaller task (or industry) oriented models that are carefully curated and trained with clean data are giving good results. That on the encoder side of life and ML for specific use cases.
General LLMs are a bluff and the amount of investment is crazy. But those are fascinating for the general public and most of the users take from granted that they speak the truth.
And that is a different discussion that has been around: should LLMs be legally bound to tell the truth?
