Scott Aaronson thinks AGI is closer

[funkervogt]

He thinks the last ten years of progress in AI exceeded his expectations. Though he doesn't give a prediction for when the first AGI will be invented, he says it's possible it could be near.

[funkervogt]

From a recent blog post of his:

Judged against where AI was 20-25 years ago, when I was a student, a dog is now holding meaningful conversations in English. And people are complaining that the dog isn’t a very eloquent orator, that it often makes grammatical errors and has to start again, that it took heroic effort to train it, and that it’s unclear how much the dog really understands.

https://scottaaronson.blog/?p=6288

[Yuli Ban]

I recall reading this, and even used that description in a discussion with someone (I don't recall if it was on this forum or Reddit or another place). But it's the truth. People want to focus on the little things that prove AI isn't fully generalized, that it's not quite transformative, that there is still a long way to go. And yet here is a talking dog, speaking in grammatically correct English and holding conversations with people. At this point, doubting what's happening is more ridiculous than the claim. 10 years ago, Gary Marcus saying that "despite the flashy show, AI has accomplished virtually nothing, and IBM Watson is an overly expensive magic trick" would've been the rational take.

I do remember using another analogy, comparing AI progress to magic. AI historically has been a bunch of technological parlor tricks, a way to fool people into thinking that computers are intelligent despite the obvious wires behind the cards. Even as late as 2018, I was saying that AI was nothing but magic tricks performed by genius programmers. The real magic was happening with the talent behind the screens who had the ability to make it seem like AI had some intelligent capability through extremely clever programming and sleight of hand. However, I also said that we'd see in the 2020s the first AIs where the real magic was happening in the computer screen, where the programming itself was doing magical things. And it seems we're there. Turns out the shortcut was large language models after all.

The fire alarm is ringing. I'd reckon it's been ringing since GPT-3, but ever since this past golden week, it's been a deafening wail.

[funkervogt]

Yuli Ban, the Turing Test will almost certainly be passed this decade, but I think your alarm over the rise of AGI is premature. I'm still not expecting it for at least 25 years, and possibly 50 years.

[Dr. Casey]

Yuli Ban, the Turing Test will almost certainly be passed this decade, but I think your alarm over the rise of AGI is premature. I'm still not expecting it for at least 25 years, and possibly 50 years.

Yuli didn't give any actual predictions for when he expects AGI...
So, with that said, when do you predict AGI, Yuli?