I think this essay brings up a great point. I would be interested in how being on the receiving end of these inquiries tends to be perceived. Are people who are asked "googleable" questions typically happy to answer? Or do they find the other person to be dumb?
I think it depends on the relationship. My girlfriend sometimes asks me things that are obvious to me and easily searchable, but I enjoy the opportunity to flex my knowledge with her. My good friend who is an excellent conversationalist asks, I’m happy to answer, because he usually has interesting follow-up thoughts. My past coworker who I don’t think is very intelligent and I wonder if we should have even hired and hasn’t improved much - I am not terribly interested in answering such a question.
This essay is good, but it misses two points.
LLMs are also used for things like a validation source,
Validation is very important! You shouldn't want all your ideas and actions validated, but most people do benefit from having their genuine virtues, efforts, and fundamental worth validated. Feeling like you matter and have genuine value is protective against depression and having someone external point out what's good about you helps you internalize it and see the good in yourself, as well as to prevail in the face of adversity.
Excessively sycophantic AIs that just validate anything are bad, but validation that actually makes people recognize what's good about themselves does make them more capable.
Ever since watching this clip of a Larry David interview, the phrase "social intercourse" has stuck with me. He uses it to describe things like dinner parties where people mix and mingle.
I thought about it the other day. I was in Cancun on vacation and was texting with a friend. I went on Google Maps to figure out what part of Cancun my hotel was in and noticed something odd: it said "Gulf of America" where I was expecting "Gulf of Mexico".
My first thought was to ask my friend whether it was a Trump thing that I missed. My second thought was that that's a lazy question: I could easily just look it up. But my third thought was that this second thought is one of a midwit.
When I put my Jedi cloak on I realize that information isn't the only consideration to think about here. Social intercourse is also a consideration. Talking to friends is nice, actually. So I asked, and it lead to a fun little exchange.
This makes me want to walk back a bit on the advice I previously had to "give it a google". Not that it's always bad advice, just that one should consider social intercourse in addition to knowledge seeking.
Consider the first example from that post; the one about poker. For me personally, I began my poker journey with such googling. It wasn't a bad idea; and yes, it's far superior to just winging it. But my cousin Jason plays poker. I could have reached out to him. He would have been able to do a fine job of teaching me, and it would have been a good excuse to catch up. We don't get to talk often enough.
So I guess what I'm trying to do here isn't really to walk back on the "give it a google" idea. It's more that I'm trying to expand and generalize it. The idea isn't to literally use Google. It's to do some form of initial research, and asking a friend (or family member or librarian or neighbor...) will often suffice.
And furthermore, I'm trying to say that when the initial research involves talking to a human, you get the benefit of social intercourse.[1] This should be part of your utilitarian calculus. Or, more pragmatically, you should probably just have a heuristic saying to "be mor social" alongside related heuristics such as "touch grass".
What about LLMs? Well, I kinda see LLMs as an extension of all of this.
Like, there's the set of things you could research by googling, and then there's the set of things you could research by prompting. The latter set is larger and is growing more quickly.
I think there's a failure mode with the first set where you don't give enough weight to social intercourse. Where you ask google how to fold your laundry instead of calling your mom. I think there are similar failure modes with the second set and that these failure modes are naturally more dangerous.
Amongst other things. Like, I dunno, maybe networking.