Pekka LundDec 4, 2022

@metzinger @anilkseth
I would love to hear your thoughts on #chatgpt .

I think it's already demonstrating the gap between "just a language model" and general intelligence is nowhere as big as many believe. And that applies to both AI and humans.

I can already imagine it accomplishing large projects, like writing complex software, by conversing with itself. Very much like we do it.

Show threadAnil SethDec 8, 2022
@pekka @metzinger having played around with chatGPT a little, I must admit I find it both impressive (for its ability to come up with at least some response at least some of the time) and underwhelming (for being very easy to trip up, and for giving zero sense of a mind behind)
Show threadPekka Lund

@anilkseth @metzinger
You have probably tested its capabilities with a good set of revealing questions.

It seems that OpenAI has purposefully taught it to state very clearly it has no mind of its own. It basically refuses to answer any questions involving feelings, opinions and such. It also claims it can't produce any original content, yet produces all sorts of stuff on request.

That leaves me wondering how convincing it could already be if it was instead taught with some kind of self-model.

Dec 8, 2022 at 3:24pmWeb
Show threadPekka LundDec 8, 2022

@anilkseth @metzinger
Impressions of course also depend on whether it's compared to earlier AIs/chatbots or humans. I think it has led to an unexpectedly fast and wide shift towards the latter. While it's not there yet, it's already close enough to make people nervous for several reasons.

Just a few years ago it would have been a big surprise if any AI could have understood and answered many of the questions correctly. Now many seem to be more surprised when it fails.