Comment by zdragnar
6 days ago
I really don't think it does, because we disagree on what the upper bound of an LLM is capable of reasoning about.
An average human may not be suitable for a given task, but a person with specialized skills will be. More than that, I believe they will continue to outperform LLMs on solving unbounded problems- i.e. those problems without an obvious, algorithmic solution.
Anything that requires brute force computation can be done by an LLM more quickly, assuming you have humans you trust to validate the output, but that's about the extent of what I'm expecting them to achieve.
Think beyond LLMs
You need to think about what comes after LLMs that look nothing like LLMs
You need to think about what robots with human capabilities, which are improving multiple times per day, is going to do.
Now add LLMs back in as your HMI