The GPT-3 hype is way too much. It’s impressive (thanks for the nice compliments!) but it still has serious weaknesses and sometimes makes very silly mistakes. AI is going to change the world, but GPT-3 is just a very early glimpse. We have a lot still to figure out.
131
724
173
6,416
It did shift for me the likelihood that we'll see something equivalent to AI passing a Turing test in a meaningful way in the next 10 years. That might be an overreaction, but it's starting to feel like the writing is on the wall.
2
1
2
36
One thing I've updated towards thinking is that the Turing test is less interesting than it seems. A big moment for me will be when AI can prove a new mathematical theorem.

Jul 19, 2020 · 7:10 PM UTC

21
37
14
529
RelevantRecentLikes
That's probably a wise position. The Turing test seems to be about it feeling profound to see such a clear reflection of ourselves in a machine, but perhaps that's not as meaningful as has been thought.
1
4
A theorem people think is interesting... "New" has surely been done already.
Aren't we making good progress on that front? arxiv.org/pdf/1904.03241.pdf
1
That might be easier than we all think. Imagine a system consisting of three parts: 1) a theorem enumerator, 2) a theorem prover, 3) a neural net trained on which theorems are beautiful. 1, 2 have been done before, all the ingredients are there. See writings.stephenwolfram.com/…
1
5
You’re validating what I’ve been saying for years: “every significant advance in AI seems impressive for, like, five minutes”
3
Didn't a machine already pass the Turing test when the participants were told they were speaking to a person for whom English was their second language. It's a narrow achievement, and it's arbitrary nature means people won't care much when it happens for real
1
Have you heard of Karl Popper?
A counter-point to that though is it does seem that as soon as a milestone is reached, and the magic dissipates as we realize an accomplishment, it feels more banal. Although maybe instead of a math theorem, it could be an original/interesting argument rather than a composition.
13
How you interpret AI's proving of a mathematical theorem will depend on whether you believe humans "created" mathematics or we "discovered" mathematics.
1
8
discovered for sure imo
2
1
20
View more replies
Oh my, #AI has suffered from such shifting of the goal posts forever! I’m still waiting for it to solve all humanities problems in less than a second (without any data) 😀
1
AI and new theorems? Done. news.sky.com/story/mathemati…
56
Oh my God, don't you read any computer science history beyond Turing? Perhaps not.
Most humans cannot do that
Ex Machina by Alex Garland mentions the issue: “In a Turing test the AI is supposed to be hidden from the examiner”. “No we’re way passed that. If you spoke on the phone she’d pass for human. The real test is: You see she’s a machine. Do you still feel she has consciousness?”
81
Is OpenAI currently trying to train a model on existing math papers in hopes to extract a new theorem that can be proved? Agree that this will be way impressive than coming up with a creative sounding poem. What I see GPT3 lacking is the amount of originality in its creativity.
What was the name of the problems of the form "the referee gave the captain a red card because he was breaking the rules. Who was breaking the rules?" Can GPT-3 do those yet?
1
Hasn’t the Turing test already been passed?
1
Load more replies