Related questions
The results of this poll are consistent with a poll of mine that (as far as I know) was never posted to Twitter or anything like that. 22 of 24 No voters as of the time I'm writing this have had an account for at least one week, so "new user accounts" doesn't explain it. It seems like Manifold's current opinion of Eliezer Yudkowsky's views on AI is neutral-to-negative on average, even among users who didn't join specifically for this poll.
My poll only has a sample size of 47 so far, but the difference between "Yudkowsky is right about AI" (9 votes) and "Yudkowsky is wrong about AI" (24 votes) is probably still statistically significant. (14 people have picked "neutral/mixed," and it seems like many (most?) think that Yudkowsky is right about risk but wrong about AGI development timelines.)
tl;dr Pretty sure this is a genuine result, y'all. "Roughly 50% of Manifold users* think Yudkowsky is wrong about AI risk" and "someone who isn't Yudkowsky is judged by Manifold to have 'won' a debate with him about AI risk" are consistent.
*this is consistent among both polls; 24/47 voters in the first poll say he's wrong about AI risk, and 7/13 in the second one say he's wrong about AI risk.
Lol! turns out you cant switch (i think)!
@Manifold ... New user accounts should not be allowed to vote in polls for at least a few days ....
there appears to have been a flood of sockpuppet accounts coming to influence the result here at the last minute.
@HenryE I didn't see DeepFates twitter post until after I commented here. It looks like that's what did it. So probably not sockpuppets, just the e/acc Twitter army pouring in.
Could the administration of the site give us to know the result without taking into account the accounts created today? ))) just for fun, I don't plan to dispute the results of Hotz's most fantastic win )))...
@Penultimate This information is available to everyone, although will take some time to get it without a script
@Penultimate I wouldn't jump into conclusion that Hotz does something to win the poll or even cares though. Maybe his funs do on their own, or maybe it's even organic.
@rpominov I saw the beginning of the poll, and it was in a completely different ratio, and then everything turned upside down; given the peculiarities of the local audience, it’s not organic ... but of course, I don’t and won’t have evidence that this is Hotz))
@rpominov The difference is only a few hundred votes? It's not unreasonable that many people would see the poll linked on twitter and click on it to vote hotz. (this was my process for ending up here anyway)
@rpominov
According to my potentially sloppy script, here are the numbers as of now.
New users votes:
Results without new users:
The hotz / yud variables contain data from the network tab of this page
(You need to click on this number for the request to appear in the network tab)
@HenryE I'd still say this is not organic if there're such tweets directed to a large pro Hotz audience, even if all their votes are honest
@Yves because they came immediately to vote for Hotz! However, I even doubt if these are bots! Unfortunately, you can't watch it here, but the voting process was completely different until Hotz caught up with voters from some of his places! This is not a gentleman's act, but he has the right to do so! He himself said that he was a cowardly tiger! ))
As far as I can tell the only major crux that came out of this was on timelines (and perhaps the relevance of timelines). There was some agreement that capable AGI systems would be an existential threat but Hotz's main claim seemed to be that was far enough away to not be a concern to us (at least not to us personally).
Manifold / Metaculus / expert consensus do not hold 100+ year timelines and Hotz' points around approaching computational limits do not seem to have moved those markets, nor did I find them very compelling.
Some priors: Hotz is a legit genius, Yudkowsky is a crank. If you stack up their accomplishments, one has made significant contributions to technology and business, while the other writes fan fiction and hangs out at a fraudulent organization that extracts money from fools. Now that you know my bias, take the following as you will.
Hotz is the winner, but not because of exceptional debate skills, he just happens to have the only grounded, rational arguments in 90 minutes. Hotz also appeared to be at a disadvantage in that he didn't want to flay his once hero. Yudkowsky is a clown that fails to make any cohesive, logical argument for his positions. His claims require extraordinary suspension of disbelief and don't build a potential bridge from where we are today to where he sees this all going.
The entire "debate" was painful to watch as the discussions were often pointless and pedantic. Yudkowsky has a strange crutch – he'll frequently fall back on external evidence (books, past accomplishments, etc) or not wanting to outsmart the audience as reasons to not have to form a coherent point. Several times Hotz wanted to have a legitimate exploration of ideas rather than a pure debate, but it became clear that this would not happen.
I wish for a better debate that explores real ideas, progress, and concepts, and ditches the mental masturbation of solving the prisoner's dilemma or turning the galaxy into spaghetti.
Until then, I wish Hotz the best as he improves his debate skills and positions. And I hope the classical Doomers elect a better representative of their party position. May all the Doomers continue to hang out on, and use the very technology they say is coming for us, for their engagement adds revenue to the organizations that fund additional AI R&D – destiny's sense of irony is boundless.
@StephenMWalkerII I wouldn't say Yudkowsky is a crank but his debate skills need more polish than his opponents. Eliezer kept making references to chess but he allowed Hotz to control the initiative throughout most of the debate. Hotz kept the tone conversational and friendly. He disarmed Yudkowsky with flattery and humor. But there's no question of who controlled the debate or who made the most compelling arguments.
George was very generous and cordial, but Yud made statements that were more logically precise. What’s more is that Hotz doesn’t fully comprehend the alignment problem. Anyone who has kids knows that they don’t align. They don’t go and kill their parents but they’re the same species with common goals. And they don’t have much power. Branches of government need checks and balances or they will become too powerful. By definition, another intelligence will have its own goals. And if it has power, by definition, there will be negative implications for others. Total destruction? I don’t know about that. But Hotz doesn’t understand why an AI should not align. “I’ve never had any tech that didn’t align” is what he said as a prediction that future AI will align as well. The basic problem was not addressed by Hotz. I don’t see how he can win this debate!