Sorry you have had such bad experiences we won't be able to convince and nobody can see the future but there are exciting things happening at an amazingly short scale.

Really? ChatGPT 3.5 and beyond models are fairly capable of understanding PoS and doing text analysis. I have never seen that issue yet with the more advanced models, although smaller/older ones tend to imagine fmthings about the text.

Last year I wrote a paper about using LLMs for definition generation for unknown words based on context, and the models did a fairly good job. https://ieeexplore.ieee.org/abstract/document/10346136/ if someone is curious.

I would like to read prompts where the models are failing in such way. The field is moving quite fast.

strong disagree. what's "long enough" to you?

i can find huge value out of it parsing natural language at even less than 1000 words. their context is way bigger than that already

I think one of the techniques underexplored in all the hype is guiding the evaluation process depending on the context. I.e. if you're generating code, it has to satisfy the parser for the given language. If the token is unsatisfactory, throw it out and try another one. Thought chains could be generated in a similar way (you can do so with special tokens, see "Recursion of Thought").

But yeah overall GenAI tends to remain hype-over-substance.

The main blocker for this is that LLMs are slow. Imagine waiting 3 seconds for your output in a pretty happy case, only for it to be invalid and have to wait an extra 3 seconds, with again non negligible chances of being wrong.

We envisioned doing this for an SQL query generator at work but with our constraints a single query already takes 15 seconds.