Also honestly requesting good use cases for *generalized* LLMs. Do you know of any? What am I not seeing?
inb4 anything code-related.
Discussion
I kinda have to have some understanding of this tech in order to advise clients and customers on how it works, but I literally cannot find good cases for this generalized code- or text generation LLMs.
I find the custom-trained Danish lexicon bot at chat.lex.dk quite useful. It is trained on paid academic experts' input, and only that, and refuses to answer when something is not in its sources. Very cool, but limited case. Very different from the claimed coming revolution.
I understand how the dopamine-fueled feedback loops work and why anthropomorphization and novelty works to lure some people in...
But too many people who I thought were competent are lured in by this, and double down when I point out the flaws in their output post-"AI".
I don't understand the difference between the (too many) people who fall for it, and the (too few) who see through it, check the output, and remain unconvinced ...
Honestly requesting perspectives on this.
@nielsa I've seen basically two types of users. 1, people who bought the hype and are trying really hard to get it to do stuff. 2, people who are marginally qualified for their position trying to use it to appear competent.
I've seen a few from group 1 slowly figuring out that it's usefulness is limited and nowhere near as useful as they thought. I've not seen that realization from group 2.
Also honestly requesting good use cases for *generalized* LLMs. Do you know of any? What am I not seeing?
inb4 anything code-related.
@nielsa LLMs can sort of do most things transformers can do (cos they're transformers), but they're much more convenient and accessible to the public
(and convenience is king)
e.g. mediocre translator, mediocre OCR, mediocre transcription
