I made a [guesstimate model](https://www.getguesstimate.com/models/15213)* with the intention of confirming my prediction of 13%, but I found that if one agrees with the community distribution on "[How many human infections of Covid-19 ... before 2021](https://www.metaculus.com/questions/3529/how-many-human-infections-of-the-2019-novel-coronavirus-covid-19-will-be-estimated-to-have-occurred-before-2021/)" and thinks that ~100 Metaculites would report a positive test result, the current median of 45% is justified. However, I think only about 50 would re...
The IARPA [Geopolitical Forecasting Challenge 2](https://www.herox.com/IARPAGFChallenge2) is starting May 15. When the first batch of questions come out tomorrow I'll see whether there are any interesting questions for Metaculus. The challenge is geared more towards finding the best algorithm to process a stream of human predictions, than it is to finding a strong team of predictors [(FAQ)](https://www.herox.com/IARPAGFChallenge2/faq). (The human predictions are from Mechanical Turk.) There will be 310 question over 6 months—about 13 per week. But is an...
I looked at how the five items on the list are doing. - The [Google trends search on "AI winter"](https://trends.google.com/trends/explore?date=today%205-y&q=%22AI%20winter%22) shows the May 27-June 2 2018 peak at 77, with a few peaks between then and now (which is to be expected given that the field of AI as a whole is growing), but nothing that indicates that discusssion of AI winter is growing. - I didn't find any keynotes with the phrase "AI winter". For NeurIPS 2019 in the list of talks I didn't find any "keynotes" but instead found "invited talks...

There are now over 100 questions in the IARPA Geoforecasting Challenge 2. I'm competing in the competition, and I've gathered some of the more interesting ones: shortlist.

— edited by datscilly

Eyeballing the [chart](http://91-divoc.com/pages/covid-visualization/?chart=countries&highlight=United%20States&show=25&y=both&scale=linear&data=deaths-daily-7&data-source=jhu&xaxis=right-12wk#countries), I initially estimated 10k. Then I looked at the results from the February COVID questions and the outcomes were better (lower cases and deaths) than the community expected (better than my final prediction expected as well). Then I read RyanBeck's prediction (which was more confident and lower than the community), and I made a more rigorous calculation b...
In 2030, will the percentage of U.S. adults that are vegetarian be <=5%? (according to a self-report poll specified in a [previous question](https://www.metaculus.com/questions/3337/what-percentage-of-us-adults-will-self-report-to-follow-a-vegetarian-diet-in-2028/)). <br/> I've noticed what I believe is a systemic bias in the community prediction for vegetarian and vegan related questions. I think for this discussion challenge it is easier to identify something for which the community is wrong about, than it is to have personal expertise in a topic suffi...

Californians will have ballots automatically mailed to them in the fall. From news about vote-by-mail in the primaries and forecaster reasoning, this could result in 10 million more Californians voting (50% -> 75%) pre eligible voters adjustment. 10 million is about 3% of the US population, which is still 3% post eligible voter adjustment, so this consideration shifted my prediction +3%.

While the United States currently enjoys a lead of almost 100 PetaFLOPS—533 PFlops vs. China's 438 PFlops in Nov 2018—as far as I know no large supercomputer projects are due to be completed in 2019 *, and this works to China's favor.

From Nov 2016 to Nov 2018 China gained 214 PFlops without completing any top 10 projects. In the same period the U.S. gained 305 PFlops, but Summit and Sierra accounts for 238 PFlops with the rest adding up to only 67 PFlops.

* China's large projects are due in 2020, and U.S.'s in 2021.

@Sylvain has picked a fairly mild one this time; the rules of this game encourage cooperation, there's little incentive to backstab, and counterplay or adversarial play to the dominant strategy is fairly weak. I think we should aim for a high as possible N early on, then lock it in place by refusing to comment on any new threads once that high N is reached. N=5 or N=6 should be possible, although I'm not committed to those numbers this early on. Let say we achieve N=5, and have 5 comment threads with 5 unique users. Then anyone can reply to all those...
Last year (in 2018) there were [15504](https://arxiv.org/search/advanced?advanced=&terms-0-operator=AND&terms-0-term=Artificial+intelligence&terms-0-field=abstract&terms-1-operator=OR&terms-1-term=Machine+learning&terms-1-field=abstract&terms-2-operator=OR&terms-2-term=Reinforcement+learning&terms-2-field=abstract&terms-3-operator=OR&terms-3-term=Neural+network&terms-3-field=abstract&terms-4-operator=OR&terms-4-term=Back+propagation&terms-4-field=abstract&terms-5-operator=OR&terms-5-term=Hidden+markov+model&terms-5-field=abstract&terms-6-operator=OR&term...

A prediction by NOAA which helped me get a sense of the timing of the solar cycle: https://www.swpc.noaa.gov/news/solar-cycle-25…

Solar cycle (wikipedia): https://en.wikipedia.org/wiki/Solar_cycle

There are now 13 matches left in the season and Fulham is 7 points behind the fourth worse team. The teams ranked 15th-17th all have 24 points with 25 matches played, and if they maintain their roughly 1 to 1 point/match ratio Fulham will have to play like the 7th ranked team to overcome its deficit (Fulham would need 13+7=20 points in 13 matches). If one of those teams play like Fulham, Fulham would need to play like the 12th ranked team to avoid relegation. Two models for a team's performance for the rest of the season could be: (1) a team is equally...
Currently there are [17 US reviews](https://www.amazon.com/Precipice-Existential-Risk-Future-Humanity/product-reviews/0316484911/ref=cm_cr_dp_d_show_all_btm?ie=UTF8&reviewerType=all_reviews) and 1 international review (and 26 ratings). In the 51 days since the question opened there's been 7 reviews; linearly extrapolating this to the whole year, there will be 47.3 reviews by the end of 2020. Taking into account how popularity might fade or grow over time, I have a main (prediction probability distribution) component at [27, 35, 47], and a half-weighted ...

On Good Judgment Open, the crowd median is 4%. I predicted 15% there and 25% here. Jgalt's latest comment is relevant and I think the Metaculus community hasn't fully updated on the news.

I checked out the big track record update from 08-27 and it's pretty awesome. I looked at accuracy vs. time for numerical questions and for both me and the community the log score smoothly, monotonically went up over time. I also looked at the calibration vs. time and found that my 0-5% surprisal bin was already at 200% halfway through the question's lifetime (5%-10% surprisal bin at ~30%), indicating that for some questions the outcome was known or guessed at before the question was half over. Also, mousing over each question in the timeline chart, it...

OpenAI blogged about their latest Dota bot:

We estimate that we used the following amounts of compute to train our various Dota systems:

  • 1v1 model: 8 petaflop/s-days
  • June 6th model: 40 petaflop/s-days
  • Aug 5th model: 190 petaflop/s-days

This compares to 1800 petaflop/s-days for AlphaGo Zero. My weak understanding is that the 190 petaflop/s-days figure is for a system trained over 2 months.

@metani It seems the survey is called the Current Population Survey and according to wikipedia it was conducted in the second week of March:

The survey asks about the employment status of each member of the household 15 years of age or older as of the calendar week containing the 12th day of the month.[4]

This question was erroneously set to close in 2021, when it actually closes in 9 days.