I fear Betteridge's law of headlines applies here.
A CS lecturer of mine told us that when he was a student he had a lecturer who advised him to be sceptical of AI revolutions. That was nearly 20 years ago. I've no doubt we'll see further steps but I'm not going to hold my breath for something transformative.
Yes, we should be skeptical of AI revolutions, but that also doesn't mean they are impossible or we should never devote some thought to updating our evaluations of the current risks.
I’d like to go on the record as being a GPT-3 skeptic. Yes, it’s a massive improvement over markov models, and yes, it will be used for propaganda. But the AI effect is very strong, and in a year or two people will be used to it and you’ll see more writing to the effect “why GPT-3 wasn’t such a big deal after all”.
Personally, my guess is that it’s actually just plagiarizing the training set in a way that most researchers will come to view as a kind of cheating. What I mean by that is, if you take some plagiarism detection software and run it on GPT-3’s output, it will ring like crazy.
I say this both because I believe it and because if it’s not the case, if we really have a proto-AGI on our hands, then being wrong won’t matter. I sincerely hope that we are a thousand years away from that, because otherwise we are plainly doomed.
What I mean by that is, if you take some plagiarism detection software and run it on GPT-3’s output, it will ring like crazy.
Somebody should try this. I ran a few paragraphs from AIDungeon through https://plagiarismdetector.net/ and got zero or low plagiarism percentages, but I'd imagine there are much better detectors that aren't publicly available.
I can't play with GPT-3 but when I play with GPT-2 I can easily trick it with counting games. It does well with 0,1,2,3,.... but things like 0,1,3,6,10, get poor responses. Is GPT-3 good at that?
Yes - reproducing fragments from various texts can look impressive, and could be useful in some applications - like creating comments on HN! (I give it a week before someone says "GPT3 has commented on HN and earned 500 Karma!!!"). But I don't think it can be a reliable problem solver or co-creator.
The fun bit is generalization. Create a pattern that hasn't been read before. Hard with GTP-3 because it's been given everything to read...
A CS lecturer of mine told us that when he was a student he had a lecturer who advised him to be sceptical of AI revolutions. That was nearly 20 years ago. I've no doubt we'll see further steps but I'm not going to hold my breath for something transformative.