that's the thing though, I have customers using genAI for ACTUAL business usecases and it's incredibly useful! When scoped correctly with an actual unsolved business problem they're trying to address. They're doing things like feeding in decades of paper contracts and extracting key fields like IF the charged rate changes, WHEN it changes, and other key fields, and building a master data record for their accounting team to go and true up all their ongoing service rates. But that whole project is like maybe $10k a month in used compute and genAI token consumption. It doesn't take that much computational power to achieve meaningful business outcomes.
But true to form, idiot executives are seeing that something works at one size so therefore it MUST work at the giga-scale, and of course it's generalizable to every problem domain.
I am also dealing with alot of goofballs doing things like feeding in large tabular datasets and asking an LLM for a trend or classification of groups inside that dataset, rather than using a traditional method like XGBoost or K-means clustering, or some other traditional machine learning technique. There's SOOOOO MUCH a trend of uninformed people using genAI to solve every problem, which is giving false business signals that every problem IS and WILL be solved by genAI. Even when there's significantly better (qualatatively and quantatatively), more mature, and CHEAPER means to achieve the same outcome.
It's madness.