I looked into doing the geometric mean this year but the issue is that it doesn't work with 0 values, and a non-trivial number of people assign a 0 probability to multiple events.
If you know of a reputable way to handle this problem that can be implemented straightforwardly in Excel I'm happy to calculate the geometric mean for you - I agree it would be interesting.
(I briefly got excited there on seeing that only one woman beat the Wisdom of Crowds and her initials were RC. Then I looked back at the answers I submitted and how incorrectly confident I was that Starmer and Reeves wouldn't last the year, that migration and income tax would go up, and that there'd be another Trump assassination attempt, and thought "hang on, that can't actually be me", and then saw my actual entry scraping in narrowly above chance!)
Good point - I'll amend the piece to add that (and see my reply below).
I also wondered if the RC was you at first, and then wondered if it was a journalist I know with those initials who got on the honour role with the Christmas Quiz, but it happened to be someone who'd also added their email to subscribe at the same time, and it was someone else. :-)
I just plugged it in and it got 0.222, which would have put it 136th. So better than the median human, but still a long way off the top performers! I'll add a note to the post about it.
I expected it to do better as I was very impressed by its reasoning (I guess doing as well as it did would have been unbelievable just five years ago, so it still is impressive!).
Thank you for doing it last year - it will be interesting to see what happens next year.
In last couple of months I notice that having a subscription to AI means being able to do detailed research in minutes that a year back the CEO of Goldman Sachs would have had to wait hours for.
who would be wear in the polls - should be where.
sadly, this [year] you did not do better than chance
I would be fascinated to know whether the geometric mean wisdom of the crowds outperformed the arithmetic mean this year (and also previous years).
Thank you!
I looked into doing the geometric mean this year but the issue is that it doesn't work with 0 values, and a non-trivial number of people assign a 0 probability to multiple events.
If you know of a reputable way to handle this problem that can be implemented straightforwardly in Excel I'm happy to calculate the geometric mean for you - I agree it would be interesting.
How did Perplexity+Deepseek do? (as mentioned in https://www.edrith.co.uk/p/what-do-people-think-will-happen-caf)
(I briefly got excited there on seeing that only one woman beat the Wisdom of Crowds and her initials were RC. Then I looked back at the answers I submitted and how incorrectly confident I was that Starmer and Reeves wouldn't last the year, that migration and income tax would go up, and that there'd be another Trump assassination attempt, and thought "hang on, that can't actually be me", and then saw my actual entry scraping in narrowly above chance!)
Good point - I'll amend the piece to add that (and see my reply below).
I also wondered if the RC was you at first, and then wondered if it was a journalist I know with those initials who got on the honour role with the Christmas Quiz, but it happened to be someone who'd also added their email to subscribe at the same time, and it was someone else. :-)
Thank you.
I think the AI entry did 0.298 so humans win :-) Might try it again this year
I just plugged it in and it got 0.222, which would have put it 136th. So better than the median human, but still a long way off the top performers! I'll add a note to the post about it.
I expected it to do better as I was very impressed by its reasoning (I guess doing as well as it did would have been unbelievable just five years ago, so it still is impressive!).
Thank you for doing it last year - it will be interesting to see what happens next year.
In last couple of months I notice that having a subscription to AI means being able to do detailed research in minutes that a year back the CEO of Goldman Sachs would have had to wait hours for.