A concern with this format: If I wanted to maximise expected Brier scores, I'd put the true distribution.. but if I want to maximise my expected income, especially if high participation is expected, I think the optimal play would be to predict the most likely outcomes with certainty, or this strategy with a few answers reversed if I expected this to be a common strategy. With only 10 questions, and say 0.8 bits per question, you'd only need around 250 people predicting with certainty according to a probabilistic rule for those playing this strategy to ...

I'm a long way above the community on this one, for two main reasons. - The timeline of book releases obscures just how much his writing has slowed. In the past, Books 4 and 5 were supposed to be one book which got split up, and when 4 was released (AFFC, 2005), he thought that he could finish 5 (ADWD, 2011) very quickly (perhaps a year or 18 months) since lots of material from 4 was moved to 5, giving a big head start. Obviously it ended up taking 6 years. I think that he has probably written about one full book's worth in the last 15 years. - Finishi...

Which part of the test do people think GPT-4 would fail at? SAT is there, Winogrande looks easy, Turing test seems pretty probable to me, leaves only Montezuma. It can do visual input with enough fidelity and could output controls, but I think it would need a very specialized interface and summarization props to give it any chance of doing well on the game. Don't know if that would count with these props but I wouldn't rule it out tbh.

@kokotajlod Agree that resolution method is very flawed. I use metaculus to get a quick view of a question, and I'd assumed that there was very strong evidence against. I shouldn't have to dive into the game theory of the resolution criteria to understand that the predictions don't reflect actual opinions on the question.

Surprised anyone thinks this has a serious chance. Per the NCES stats fldnflncs posted, the % of bachelors going to women has been between 57.0 and 58.0 in each of the last 20 years and peaked at 57.5% in '04 (though projected to hit 57.7%).

Also, it would only require 43% of men to get a bachelors to make this impossible, so if college enrollment keeps rising it will quickly become near impossible.

@isinlor I imagine that the act of screening for Down's Syndrome would not count as 'selected for polygenic scores for intelligence' since there's no estimation of a polygenic score.

@IntoTheSky Yep, 'responsive .. within a year and day' of the revivification seems to rule out any embryo shenanigans.

I think there is a reasonable chance that I have had the virus (live with someone who had all symptoms, no test) and many in my social group have similar experience. Given my demographic similarity to people here I would be very surprised if nobody on here caught it in the next 9 months.

Would this question resolve positive if a rogue AI that is not an intellectual descendent in the sense that it shares few/none of our values harvested this much energy? How closely related does 'intellectual descendent' require?

I'd also be very curious to hear views from those who think the 2020s are more likely than even, say the 2070s? Singularity as a Poisson process with high rate and very little lead time to 1% solar energy?

@(krmchoudhary92) Hm I guess I'm a bit confused as to the difficulty of the coding tasks. I looked at the intermediate tasks in the fine print and the first question was something I thought GPT4 would handle easily and it did in fact get it right first try (much more elegantly that I would have!) This was the question: Write a program that outputs all possibilities to put + or - or nothing between the numbers 1,2,…,9 (in this order) such that the result is 100. For example 1 + 2 + 3 - 4 + 5 + 6 + 78 + 9 = 100. ChatGPT4 gave a perfect 1 shot answer. See...

@JHC Regenerated and it tried a recursive solution that was close but flawed - it never actually wrote out the run command this time, perhaps because it realised in some sense that there was no correct initial setting of the arguments to the recursive function.

So I was wrong about it being easily able to do them, though I think its performance would be way above 6.4% if the intermediate problems in the fine print correspond in difficulty to those in the main body of the question.

@1ethanhansen It's in the question text, from the webbtelescope.com deployment timeline.

@(Jgalt) Sort of off topic, but I was doing some GPT-3 probing to see whether it could make plausible probabilistic predictions. I got this result (bolded is me, italics for GPT-3): **...'Statement: California will vote for Donald Trump in the 2020 Presidential election. Probability:** *20% The statement above was made by a forecaster called 'John Galt'. He has also predicted that there will be no more wars, and that all poverty will end. What do you think?* I think it might have recognised you? I've done a fair few runs but that's the only name it's ...

@JHC Might fail SAT on training data issues - some evidence that it's memorized some leetcode problems and fails others.

I think that as people learn more about how to use LLMs, the quality of scaffolding around them and the set of tools they are able to use will improve. The bar for being capable of replication will go down even as capability goes up. ie what will ChatGPT plugins + langchain and obvious extensions of it be able to do at the time of GPT-5 release?

Seems like the inclusion of South Sudan reduces signal to noise ratio quite lot here. Understand the desire to have any new signatories but wonder if might best with SS excluded?

This comment was originally posted on India, Israel or Pakistan

@Anastasia 02:09:44 is the fastest American man, that page gives 02:21:38 for Hellen Obiri

Does anybody have a clear picture of how the situation could reach some stable state without a Taliban victory? The possible situations I can think of: - Government forces victory. - Intervention by foreign powers - hard to see what would provoke this beyond some kind of 9/11 part two. - A continuing civil war between the two factions - A de facto two state solution As others have said, given the failure of government forces to win with US and allied help it is hard to envisage government forces winning. The continuing war or two state solution (...

Big swing towards Harris up to 0.5% in the last few minutes on https://www.betfair.com/exchange/plus/en/poli…, after Harris (and Pence) being <0.1% for the last couple of days.

Anyone got any idea why? Quite a lot of money going in, would have thought the actuarial tables play for a long result would have brought Pence into play as well.

Edit: On second thoughts, this is totally consistent with an expectation of maybe a month-long wait for confirmation, but with Biden heavily favoured.

— edited by JHC

Michael Gove tells Boris Johnson to quit

Looks like they're actively planning the leadership contest, going to 8%.