Creating a new endpoint that accepts data as sent by the AI-generated code is very different from accepting AI-generated errors for existing endpoints. You're talking about the latter.

That chart says it all. From 10% of industrial companies making losses to 30% in less than 15 years.

Go on...

EDIT: Found this: https://github.com/HigherOrderCO/HVM1/blob/master/guide/HOW....

A reference would be nice

> I'm a Data Scientist currently consulting for a project in the Real Estate space (utilizing LLMs).

Consultants are obviously making huge amounts of money implementing LLMs for companies. The question is whether the company profits from it afterwards.

Time will tell, but I would cautiously say say yes.

Note that I don't usually work in that particular space (I prefer simple solutions and don't follow the hype), didn't sell myself using 'AI' (I was referred), and also would always tell a client if I believe there isn't much sense in a particular ask.

This particular project really uniquely benefits from this technology and would be much harder, if possible at all, otherwise.

> Every one I've spot checked right now has been correct, and I might write another checker to scan the results just in case.

If you already have the answers to verify the LLM output against why not just use those to begin with?

Not GP, but I would imagine "another checker to scan the results" would be another NN classifier.

Thinking being that you'd compare outputs of the two, and under assumption of the results being statistically independent from each other and of similar quality, say 1% difference between the two in said comparison, would suggest ~ 0.5% error rate from "ground truth".

Maybe their problem is using LLM to solve f:X→Y, where the validation, V:{X,Y}→{0,1}, is trivial to compute?

What's the expected effort needed for supporting GPUs other than Nvidia — e.g. AMD GPUs or the GPU in a MacBook Pro M1/2/3?

As I understand, it's a lot of work because there's no common way to target these different GPUs. Is this correctly understood?

Didn't the model ever fail to generate realistic-looking content?

If I don't know better I'd think you just cherry-picked the prompts with the best-looking results.

What you see there is a product, not the scientific contribution behind it. Consequently, you see marketing material, not a scientific evaluation.

Unfortunately also the majority of scientific papers for eg. image generation have had completely cherry-picked examples for a long time now.