Fermi Questions (fermiquestions.org) is a Wordle-like game where you try to guess the answer to Fermi estimation questions with 6 or less tries. After each guess, you'll see if your answer was too high or too low. You win if your guess is within ±20% of the correct answer.
Example questions:
- How many new cars were sold in the US in 2024?
- How many humans have ever lived (including those currently alive)?
- How many chickens are slaughtered for meat every year?
The skill of Fermi estimation is also extremely useful if you want to perform better here on Manifold. As Philip Tetlock shows in his book Superforecasting, many of the best forecasters break down complex questions into smaller, more manageable components which is exactly what you can practice when playing the game.
In the first 24 hours after going live it had some 1,100 unique visitors, which primarily came from Hacker News. Resolution source is the unique visitors count on the publicly available analytics dashboard: https://plausible.io/fermiquestions.org
Update 2025-08-04 (PST) (AI summary of creator comment): In addition to the default bot-filtering by the Plausible analytics tool, the creator reserves the right to manually exclude any suspicious activity from the final unique visitor count.
Update 2025-08-04 (PST) (AI summary of creator comment): A unique visitor is defined as a distinct visitor to the website within a 24-hour period, based on their IP address. A user who visits on different days will be counted as a new unique visitor for each day.
I've greatly appreciated people here sharing screenshots of their game statistics; please continue to do so! This is great fun!
Another clarification that might be important: For the purpose of this market a unique visitor is defined as a distinct visitor to the website within a 24-hour period, based on their IP address. So if the same user returns on a different day, they are counted by the analytics dashboard as a new unique visitor.
@TheAllMemeingEye If you visit the site from the same browser, it should show your previous progress. Maybe you have initially played the game by opening the link here on Manifold, and now you’ve looked it up in a different browser? Anyways, at some point I’ll also allow the creation of accounts so that we don’t have to rely only on using local storage.
@ms By tomorrow I should have some tooltips implemented, and I agree that this can be confusing. But kudos to you for playing 3 questions without ever seeing the arrows :)
You can now see an explanation for what the arrows mean in a tooltip. The tooltip is visible by hovering, but it is also shown automatically when you are playing the game for the first time ever and have made an incorrect guess.
lol nice (SPOILERS)
I guessed there’s maybe something like 1m doctors in the US, and they spend maybe 10% of their professional life’s studying
SPOILERS
😎
@ms Even your assumption that there are 1M doctors in the US is exactly right because it literally is 1,01M. It’s actually one of the hints that you would have seen if you had 2 incorrect guesses. My approach for that question was basically to think about how many doctors are there, how many years do they work, how many students does it need to replace those retiring, …
One of the main downsides of the game so far was the focus on doing somewhat trivial binary search if one did not succeed in getting within ±20% of the correct answer with the first guess.
I think I found a solution to bring the focus a bit away from the binary search and prompt people to rethink the original question once more, but now in light of new information: showing a hint after the second incorrect guess. For example the hint "The US covers 1.87% of the Earth's surface." is displayed for the question about what percentage of the Earth's surface is land.
I hope that this brings into the game a whole new dimension where you have a second moment after the initial Fermi estimation to really think through your guess and the assumptions you have made. How does the new information received through the hint impact your priors?
I think those things put together now make the game a very compelling training ground for getting better at Fermi estimation and updating your beliefs in light of new information without over or under-reacting.
I would greatly appreciate feedback from all of you.
Idea I just thought of:
Global/regional leaderboards of running average number of guesses needed and/or running average log closeness of first guess to correct answer (maybe subtracted the standard error so that you also need to have achieved it over many games), to trigger competitiveness in players
@TheAllMemeingEye I like the idea, but I just don't know if a leaderboard can work on something as easy to cheat at as this
@MingCat what if it was like Impossible Quiz and you get instant game over if you switch tabs? Sure you could use a separate device but it's much less convenient
@MingCat alternatively, maybe percentile rather than full leaderboard could work better since less incentive to cheat for social status, like in ethnoguessr but running average rather than just today's score:
In this distribution it looks like out of about 3000 players there were about 10 possible cheaters, but they barely affect the overall percentiles
@TheAllMemeingEye I like that a lot. I’ll probably not get to that this week, but I’ll prioritize it next week. The game really needs some way for players to see how they compare to others.
