@ugandamaximum I don't see any evidence that his bets are promotional. I'm also confused why people say that his wagers are trivially small. Most are $100, which is pretty normal for casual bets. (Do you need people to make multi-thousand dollar bets for them to look "serious"? If so, then perhaps Metaculus is not the site for you, since everyone here is betting $0.)

@Jgalt Can you explain how this is "virtually inevitable"? My understanding is that there has been essentially no progress in extending maximum human lifespan during the whole course of human history.

@(Jgalt) I agree with the weaker claim that progress in the last century or two has been very fast compared to all prior time in human history. I disagree with the strong claim of inevitability and here's why: * Extending maximum human lifespan seems really hard, as there are currently no medical therapies that have been shown to work in any non-trivial degree, and that's not for lack of trying -- people have been taking drugs and random supplements for decades with pretty much no results as far as we know. Aging is a deeply rooted biological process an...

I agree with the current community prediction. [Jeff Wu et al.](https://arxiv.org/pdf/2109.10862.pdf) last year produced a model that was sometimes capable of producing coherent summaries of entire books, > Our main result is a model that can be applied recursively to generate plausible summaries of entire books. Our approach lets us summarize books of arbitrary length – we achieve believable summaries on books with hundreds of thousands of words by recursing to depth 3. With a non-recursive approach, generating or evaluating a book summary requires a h...

Does anyone want to re-launch this question with a smaller range?

@Uncle Jeff "The academia-loaded Metaculus community is severely biased toward Democrats"

This is a testable hypothesis. I look forward to testing whether this is actually true... eventually.

@(literallytrevor) It might be worth inviting someone who has a lot of experience playing Diplomacy into this thread. I don't really understand why it would be a major disadvantage for the AI to be perceived as an AI. People will presumably simply take that information into account, and modify their strategies based on their understanding of how the AI operates. I also don't think it's extremely difficult to have an AI generate long-term board game strategy, or produce text persuading people that their strategy is good. Presumably, the long-term strateg...

I'm confused here at the low probability. I predicted it would be higher. Though in hindsight I probably should have conditioned on no existential risk. Another thing I would guess is that people may be overestimating the likelihood of a false positive ie. a therapy is developed which extends lifespan but isn't a rejuvenation therapy. As I have pointed out in another comment here, I am skeptical that even a full cure to both cancer and heart disease would trigger this question to resolve non-ambiguously.

@(Anthony) @Gaia Here's an example setup of a yearly adversarial contest that I'd be particularly excited by: There are two sides to the contest: a human side and an AI side. There are also two phases of the contest. The human side of the contest is represented by eligible participants who, during the first phase, submit question-answer pairs. Eligibility could be open to anyone with a Metaculus account, though if the contest is flooded with spam or low-quality responses we could restrict eligibility further. Each participant who submits question-answe...

@randallburns "There is a high profile Matthew Barnett that is different i suspect from the one in the question."

Is this a joke?

I'm Matthew Barnett, the one who wrote the question. We could just say "Metaculite Matthew Barnett" if needed.

I'm also not sure we need a bio for Robin Hanson. But I'll let other people give their thoughts about these matters.

@A_Guy I'll believe it when I see the lifespan gains. I'm not sure how much to trust aging reversal as measured by an epigenetic clock.

Here is a source that compiled a lot of statistics to compare James Holzhauer's performance with Ken Jennings.

@TeeJayKay I think in that case it should resolve ambiguously.

@stupidme I think Guam should count, and already does under the current wording. In WW2, Japan attacked Hawaii, which brought us into the war. Yet, at the time, Hawaii was not a state.

@(jacyanthis) Thanks for this reply. I agree with many of your points, and disagree with some others. > Concern #1 (Overfitting): While I would agree that these are five of the best tests one could specify using today’s benchmarks literature, any set of five tests out of the thousands imaginable will be susceptible to overfitting. The model may perform very well on these tests but very poorly on other intellectual tasks, such as a robotics task of navigating an obstacle course or a coding task of building an entire website with certain specifications. ...

@DavidWayne It resolves in 2010 USD, so it doesn’t matter how much the USD inflates or deflates.

@(kokotajlod) "If this is so, (1) isn't that a calculation for superhuman performance, not human level?" I'm not particularly attached to the term "human-level". I agree that such a model would be narrowly superhuman, as no single human would be capable of doing all of what it does. However, at the same time, none of its abilities would be beyond what individual but separate humans could write (as the model would literally only predict actual human text). "(2) It would be really cool to see the calculations done by parameter count or training-compute-...

A new submission from the T5 Team at Google scores 88.9, improving on the previous best of 84.6.

— edited by Matthew_Barnett

@Sylvain Here is a survey that would qualify. I'll add it to the description. It turns out only 6% reportedly subscribed to the Everett interpretation. Perhaps this question will only resolve positively in other worlds...

Given that DeepMind's recent weak AGI was based on deep learning, I don't see any reason to think that this question will resolve negatively. Progress in deep learning could suddenly halt, and another paradigm could take over in its place, but that seems increasingly unlikely.