@james.jj.barry Agreed. It looks like the the 5th percentile for the current forecast (which is part of the report) is at approximately 1,470 deaths, which seems quite low, given that there are currently 1,328 deaths. The 142 deaths required for this in the next 18 days is less than the 172 which have occurred in the last 5 days.

Update: 1,405 deaths have now been reported on http://ldh.la.gov/Coronavirus/ (COVID Tracker's source). The community 5th percentile is still in the high 1,400s.

— edited by elifland

YouGov: When it comes to how registered voters plan to vote...

• 39% plan to vote in person on election day
• 19% plan to vote in person before the election
• 39% plan to vote by mail

(I copied this comment from https://www.metaculus.com/questions/5053/what…)

My forecast is much more bullish than the community on hitting the 20k mark relatively soon. Maybe this is due to people not updating on the 17,000 mark having been broken in June?

Looks like there are currently 14 pledger/member athletes. The 18% the community is currently giving to <20 at the end of 2021 feels way too low.

Thanks @borromini for the updates! Based on linear extrapolation, I get 105 reviews, but the rate of new comments seems to be speeding up, at least in July. My Elicit distribution assumes that there's a 2/3 chance it ends up above 105 reviews and no chance it ends up under 49 (the current amount).

[I work on Elicit.]

— edited by elifland

I used Elicit to create a distribution for this question. I ended up more peaky than the community in the August-September range, and was able to assign little probability to dates before the July 17 re-opening date to be postponed.

[Disclaimer: I work on Elicit]

@Natalia_Mendonca I think you’re right. I incorrectly thought all questions closed at 6 PM EST.

@casens You can download the complete CSV data at https://covid19.apple.com/mobility. The Github data is out of date for some reason.

My forecast attempts to balance the recent increasing trend with the longer-lasting decreasing trend.

@Anthony That is yesterday's number, you can see the date as the first column in the spreadsheet. It's more clear here which I think mimics the spreadsheet? https://covidtracking.com/data/us-daily

ESPN: NBA, NBPA announce playoffs to resume Saturday, new initiatives

The NBA and National Basketball Players Association released a joint statement Friday announcing that the NBA playoffs will resume Saturday and that the league and its players will work together on several initiatives to promote voting access, combat social injustice and racial inequality, and advocate for police reform.

@krtnu, thanks for the historical data. Based on the data, I created an Elicit distribution based on a rough extrapolation + uncertainty.

[I work on Elicit.]

I'm confused why the community prediction is so high, given that if I'm reading the data correctly the current value is ~50 and has been relatively stable for almost a month, albeit on a slight upward trend. My Elicit forecast has a lower median and less uncertainty.

[I work on Elicit.]

— edited by elifland

@(Anthony) > The experts here are very talented, smart, and capable people. I expect that they are rapidly getting better at general-purpose predictions through this process, which is great! I hope that you are right as having both subject matter expertise and good probabilistic intuitions/reasoning skills would be the best of both worlds for forecasting! With that being said, the experts have making a similar mistake several times in a row as pointed out on the [FLI Podcast with GJP superforecaster Robert de Neufville](https://futureoflife.org/2020/04...

The community seems way too optimistic here. The best few-shot submission on the leaderboard is at 76.6, pretty far away from the 89.8 human baseline and having closed ~1/4 of the gap in the ~1.5 years since GPT-3. Naive extrapolation would puts this at ~2026 and this also seems about right to me intuitively.

I'm surprised by the [superforecasters' current prediction on a similar question](https://goodjudgment.io/superforecasts/#1439): > Between 21 April 2021 and 30 September 2021, what will be the highest seven-day median of daily new confirmed cases of COVID-19 in the US? > > 1. Fewer than 90,000: 25% > 2. Between 90,000 and 140,000, inclusive: 61% > 3. More than 140,000 but fewer than 210,000: 12% > 4. Between 210,000 and 300,000, inclusive: 1% > 5. More than 300,000: 1% The aggregate seems very confident that it won't be >210,000. In the past...

Does anyone have the base rates handy for how often standout charities have become top charities in the past?

I vaguely remember someone commenting them on another ACE-related question.

@stuhlmueller Another related question, albeit short term: Will an AI be trained using an Avogadro's number of operations before the end of 2020?

— edited by elifland

@casens Thanks for posting! Using Elicit, I turned these beliefs (slightly changed from 2 days ago) into a distribution for submission.

I'm surprised the community median of 52M is so low. There are currently ~30M cases and ~300k new cases per day. With 104 days left in the year, we'd have 61.2M with the current growth rate.