[nerdjon]

Given the current idea of LLM's still struggle with confidently giving wrong information, all of these products are seriously misguided. (It doesn't help that "AI" has become such a blanket term in the last year or so that does "AI" mean "LLM" or more traditional purpose built AI).

It isn't clear to me that given the nature of LLM's can we actually solve this problem. It isn't thinking critically and never will without it being a different tech. (Someone correct me if I am wrong, but it seems like this is just a fundamental problem with this type of tech).

There have been very very few actual use cases of what we are now calling "AI" that actually seem to provide any real benefit. The only one that I find myself using on a daily basis is helping look through and summarize my personal notes. Something that the nature of an LLM is ver well suited for.

[gwervc]

I agree. LLM almost-solve a traditional NLP task which text generation. Calling that intelligence gives the wrong idea about the tech application. That's no reasoning in generated text, what is amazing is that the output is linguistically correct.

Prolog, ontologies, computer vision, deep learning, classifiers, etc. all have been called AI despite being very different things and their inherent limitations. At this point AI is just a label thrown at the newest cool tech.

[nerdjon]

> what is amazing is that the output is linguistically correct.

Right, and I am not being critical of the technology to diminish this achievement. It is a great achievement.

But it feels like we have moved so far past it being "just" an "LLM" to already considering it a general purpose AI when it just simply isn't.

The problem is it fakes it enough, in enough situations, that a lot of people seem to come to the conclusion it is.