2 ms·

So you use it for generation where validation is easy. If validation were harder, the calculus would change. I also get a lot of mileage out of it, but it is i

by esafak 2y ago

So you use it for generation where validation is easy. If validation were harder, the calculus would change. I also get a lot of mileage out of it, but it is important to recognize its shortcomings.

Kuinox 2y ago
of course, I don't try to replace my whole job out of it. ChatGPT is 'good' at doing statistical analysis with python on a given dataset, that can help in the harder task you quoted "distinguishing between ideas that are correct, and ideas that are plausible-sounding but wrong". The job itself cannot be easily verified, but you can use LLM on a subset of theses tasks.
mewpmewp2 2y ago
There are plenty of use cases where validation is easy.