I don't think it's necessarily as hard as you think. Can't do math? Teach the model to use an external mathematical expression evaluator to do math. Makes up inaccurate information? Teach it to query an external knowledge base. How hard to implement do you think that is? They're already working on it.

> How hard to implement do you think that is?

Very, very, very, very hard.

It's not that hard:

https://github.com/williamcotton/empirical-philosophy/blob/m...

@williamcotton - saw this the other day in the prompt engineering post - I enjoyed it. Sent you a message on your site but something looks broken. If you use linkedin I requested to connect.

Doh, I’ll look into that contact form! I’ll check my LinkedIn now. I’m glad you found it interesting!

I think there is a greater tolerance for error here though.

Cars haven't made it yet because the assumption about allowable errors was wrong. Turns out it isn't "good enough" for self-driving cars to be safer than humans.