LLMs are pretty bad at what they’re used for. They’re inherently limited and flawed because they rely on a false premise - that by „understanding“ language you get intelligence as a result. Language is a product of intelligence - a consequence of wanting to communicate with others. If you’ve ever struggled to „write down your thoughts on paper“ you’ll no there‘s no way it captures all the complexity of your mind.
Well-scoped machine learning models are far more useful in their specialty areas but there’s way less interest in those.
There’s a very limited number of things I feel like LLMs are good for, and even then they’re not worth the environmental and financial costs.