Thread with 10 posts
jump to expanded post
LLM chatbots are very convenient and useful but they are at the end of the day not actually omniscient computers. they are a computer simulation of “some guy”. you are just asking Some Guy all your questions. he is at best an unreliable secondary or tertiary source
screenshots of chatgpt or whatever are the equivalent of “my buddy eric said”. they are not a replacement for having a personality and a mind of your own
if someone goes around doing nothing but mindlessly quoting things “some guy” said, even if that someone is really cool, eventually you start to see through it and they just seem pathetic? same phenomenon
my buddy hanekawa knows everything (she protests that this is untrue and that she only knows what she knows)
i know people will read this as me shitting on LLM chatbots, but i think it is a useful way to look at them overall? like, wow, we invented Some Guy who lives in your computer and has infinite patience for answering questions. that's kinda cool, kinda handy. but, still, Some Guy
some guy who has read every single word posted on reddit, with predictable results
a while back a friend was trying to learn programming and i dedicated a lot of my time to giving her advice and answering her many, many, many questions along the way. and that felt quite good, and i was glad i could help her, but i started to worry i might be helping too much
there's a line there and i'm not sure where it is, but this is certain: if at every single step in learning something, you just ask someone else to think for you — in other words, if your brain never has to work for it — you simply Do Not Actually Learn Anything. period
i'm not sure what all four horsemen of the human thought apocalypse are, but chatgpt (making your brain not have to work for it) and social media/smartphones (constantly distracting you, so don't have the attention span needed to work for it) must be two of them…
@hikari This matches with my view—Back 15 or so years ago when over reliance on Wikipedia was the problem of the day in academic work, I remember a colleague described it as being as trustworthy as “some guy down the pub”. Can’t be cited directly, shouldn’t be trusted as the only source, but may or may not be knowledgeable and able to point you to more reliable sources to verify. I’ve thought of LLMs in the same way for a while now.