It isn’t just one thing. The big money wants to present this unified front to the public like LLMs are a single commodity anyone can use. In reality they’re a collection of complex tools that few can use " correctly" and whose utility is highly specialized for niches those few find valuable.
So you’re correct in a way. I’m sure model decoherence isn’t helping much either and isn’t as visible in those niche applications as it is for the general public.
I swear AI has been getting worst. Probably from all the AI content it’s been learning from.
As use has been scaling up, the big companies try to use smaller and cheaper models to save money.
It isn’t just one thing. The big money wants to present this unified front to the public like LLMs are a single commodity anyone can use. In reality they’re a collection of complex tools that few can use " correctly" and whose utility is highly specialized for niches those few find valuable.
So you’re correct in a way. I’m sure model decoherence isn’t helping much either and isn’t as visible in those niche applications as it is for the general public.