In your opinion, what's the key difference that makes AI lead to world becoming stupider compared to other technologies that lead to more widespread knowledge, such as writing, book printing, calculators, computers and Internet?
I think you’re vastly overestimating the amount of work people either want to do or pay for. A worse version that’s nearly free or instant will win regardless of limitations the majority of the time.
The comment was about "you're not going to know anything because the LLM is doing it for you" which is easily obviously true. This won't stop anyone from getting a PhD or a high salary. It will just stop them from knowing things, and possibly being able to build cool stuff (specifically in the example of software dev), though even then you could argue that a good working knowledge of how to prompt an LLM is roughly as performant as a good working knowledge of a software language.
Not obviously true to me. LLMs don’t know everything, so they can’t solve all problems. You can try and give it the proper context but you still need to understand the context is necessary, which requires knowledge.
Kids who grow up today will understand and intuit that synthesizing 3 facts from an article doesn’t constitute “work” any more than multiplying 4-digit numbers. Adults will eventually pick up on this as well.
I've seen this in coding, where as an experienced coder I can spot where the LLM is doing bad things, but if I try using it in a language or environment I'm not familiar with, I get lots of errors that I don't understand how to fix and all I can do is feed back the errors to the LLM and hope it does better next time.
So I guess my point is that if we end up using LLMs for everything then we have a chicken & egg problem - LLMs don't work for beginners, but beginners don't learn anything while they're using LLMs because they can't get past basic errors.
Your point about multiplying 4-digit numbers would be valid if calculators often made basic maths mistakes, and you can only really use them if you already know the approximate answer so you can detect when they've made a mistake.