@AlejandroZerboni wrote:


I do not think that word means what you think it means.

Any thoughts? I'd lean 'yes.'

@randallburns Anyone know what the baseline for this question is? It occurs to me that (especially if "better off" is interpreted financially) most people tend to be better off as they age. So I'm not sure how to interpret this number.

I guess we should not be surprised that there seems to be a high prevalence in well-known figures, as they preferentially are the type to do a lot of international travel, meetings with people from many locations, etc. There's probably an adjustment factor from "typical person" to "well-known public figure" that would be interesting but pretty hard to estimate.

Very depressingly, the community prediction for this question has gone up a factor of 288 since March 1 until March 28, about 23% per day, which is just about how fast the cast numbers have been growing...

@ignorance.prior This feature is about to go live! Stay tuned a few days...

@bzial Also, we'll soon have log scores assigned for numerical questions (Briers are not really defined), so that may help a bit.

Commenting here so this does not get lost in the election frakas. A nice new update is live, including private question sharing, and much improved search.

Can I comment that this is like a really, really crazy number of papers? This does not seem like an exponential that can really go on for that long. Is there any way of determining how many of these papers come from academic institutions and how many from industry. The former, in particular, has some pretty stringent limits as to how fast it can grow: hiring faculty is pretty slow and there is a real shortage of candidates. If it's mostly industry that can scale up faster, though you still might imagine running out of people if the number really has ...

@James wrote:

@Anthony - the real-life bet this question is based on should have resolved now. I am guessing negatively?

Apparently Huw and Carl do not agree so it's now up to Max, Martin & I, which I had really hoped would not be the case. :-O

In correspondence with them, more info. soon(ish).

Thanks! My favorite fail:

Q: What is the answer to "Add 34957 to 70764"?

A: 34957 to 70764 = Ah yes, 34957 is a type of cheese that is made from sheep’s milk. It’s not very popular though.

Agreed. Sadly, I put the destruction of human civilization at substantially higher than 1% probability. OTOH I think if we make it to 2069 the probability is quite high that we'll put a person on mars.

Flagging a new update as of Feb 13, 2020.


While I still think the spirit of my meta-prediction has come true (witness the many, many articles expressing that even if models give 5-10%, they may still be too low, Trump can still win, etc.) I concede that my number was too high, and I've paid my 92 points for it ;-)

@notany That's right! But what is the definition of AGI? What I'm trying to do here is operationalize an AGI that seems virtually certain to be able to radically self-improve, even if it might lack other qualities and capabilities. I'm not a fan of the "AGI is what can do anything a human can do but better." There are other definitions it might be interesting to have questions about, e.g. what will be the first agent that can score 90% on the Winograd schema *and* pass a stringent Turing test *and* play a new game approximately as well as a human, et...

@steven0461 Per the track record, over the past two years the Metaculus prediction has been slightly better on both binary and continuous questions. It's relatively recently been updated, and is probably even better now, though that will have to be proven by actual results and not just backtesting.

It's likely that we're going to split off a coronavirus-focused domain since these are proliferating so much (which is great!). Keep in mind Metaculus is running several projects, which led to a spate of animal welfare questions, then climate change/environment. And there will be more such clusters down the pike.

Like the question about the end of the Universe (and the eclipse in 2500, among others), our hope is that people will make good predictions on this out of fun rather than a strict point motivation.

Or, you may want to bequeath your Metaculus status to your offspring or others; if you write this into your will then you can forecast an accurate probability without any issue!

There are existing track record compilations by Rodney Brooks (eg this one) and (other end of the spectrum) Ray Kurzweil (e.g. this one) that could be at least partly mapped to existing AI questions and would be fun and useful to import and extend.