164 Comments

Scott, would you consider redacting my name from this article? I'm really tried of it being constantly brought up everywhere, and I don't need more of my acquaintances asking me about it. (I don't mind if you include a link for people who want to click through, I just don't want my name broadcast to every casual reader.) Especially considering the very negative spin you're putting on my actions, and the details you get wrong, like claiming that this was my life savings. Obviously my actions were public and I don't think there's any moral imperative that you don't talk about public actions, but I'd appreciate it.

Expand full comment

Maybe I'm misunderstanding but if the claim regarding long term forecasting is merely that experts can do better than random members of the public or than just looking at the base rate then why would anyone expect it could have been false? Surely the skeptics must have something less extreme in mind.

Surely it shouldn't be a surprise that you can improve prediction accuracy by obeying the probability axioms (I bet u can even prove some kind of domination result about this) which is one way experts did better. And even if we restrict our attention to assignments that obey probability axioms it would be crazy if the mere passage of time eliminated any effect of inside knowledge.

I doubt that even the most extreme skeptic of long term prediction would claim that there are no generalizations that hold beyond 25 years. For an extreme example, an expert might know that SR seems to forbid FTL travel and if non-experts didn't then you'd expect that expert to have an edge in predictions about interplanetary travel.

So one thing experts might do is say: hey I know that long range predictions are really hard so I'll just use base rates or other competitive method with the minor tweak that I'll insist my model obey probability axioms and give a very high probability to existing laws of physics remaining true. Surely that can't be enough to count as saying long term expert prediction works.

Expand full comment

Regarding Balaji, what's any more puzzling about his behavior than that of anyone else saying wrong/incoherent things on Twitter?

I suspect we tend to see large monetary losses and think: damn you'd have to be really confident/careful/whatever to risk that kind of amount.

But given how rich he is and the rapidly diminishing returns of money to utility Balaji is probably being less irrational in wasting that money than I am in wasting my time with this comment.

Same with Musk. Losing billions on Twitter might seem crazy to you and I but is it really than irrational for the very rich to try and turn money into social status/feeling of pwning someone on Twitter like the rest of us do with time (also sometimes refusing to back down after saying dumb shit)?

Expand full comment

Balaji strikes me as conflating several different predictions that, while not independent, have their own drivers and probabilities.

Hyperinflation needs to occur. People need to see BTC as a viable inflation hedge and purchase it in response to hyperinflation. The level of hyperinflation and the price impact on BTC needs to drive the price up about 30 - 50x.

Feels like the prediction market questions you describe above are a bit less contingent on a series of probabilities.

Expand full comment

I think it is hard to make a good question about the moat claims. I'm pretty sure the memo assumes Google and OpenAI have an advantage to train models at GPT-4 scale if advantage is measured by GPT-4 rules. And the threatening open-source models won't need to be nominally better than GPT-3.5 at GPT-3.5 things… the question is whether a slightly weakened GPT3-level thing compressed into a smartphone-local reduced version and given interfaces with a radically different structure would be much more useful.

… and to whom! Maybe polish and UI work will still be a moat for selling to corporations, it's just the stuff for doing novel information processing work that will depend on wrangling the infinitely forked open-source systems. (Like with OSes: no real troubling selling Windows even if new sofftware ends up clientt-server with web servers running Linux or maaaybe FreeBSD and web browsers being fine on whatever)

Expand full comment
May 23, 2023·edited May 23, 2023

Hello there, I'm a cofounder of Manifold!

The Whales vs. Minnows market was originally quite a fun and creative battle, despite the way it ended.

- More than a hundred derivative markets were created: https://manifold.markets/group/whales-vs-minnows

- A user used AI to compose an epic saga telling the Whales vs Minnows story, voiced by David Attenborough, which is brilliant: https://www.youtube.com/watch?v=qV3bgODqIBc

- Isaac had a big reveal where he had hundreds of NO bettor accounts sell out of their position and thus flip the advantage to the YES side. (Before succumbing to an influx of new minnows from Reddit.)

- A council of top Manifold users formed to help govern the resolution decision to make sure it was fair. They agreed to resolve the market at a random time — when next bitcoin block was mined with a hash ending in "00".

All this speaks of a very passionate community including at least a thousand people directly participating for fun. Many of our users are also learning the mechanics of prediction markets in the meantime – for example, understanding how the price of shares relates to the probability, or creating their first limit order.

I think Manifold was able to learn from this market too and implement a mana-purchasing limit to prevent future problems like this one. It's also worth mentioning that we've been operating for a year and a half and this is a pretty unique event.

Aside from all that, we do still have exciting plans to make prediction markets better and more useful for real-world forecasting!

For example, just today we asked if a cool new mechanism we are working on will be the most impactful feature we build in 2023 and it's currently at 54%! https://manifold.markets/Austin/will-multi-binary-be-the-most-impac

We've also identified a clear path to monetization that is better than relying on users dropping $29k to participate in meme markets.

We have seen that prediction markets produce valuable information that answers users' questions.

So, what if you could spend mana to get the wisdom of the crowd to work for you? You would be buying knowledge from other users.

Now you can do just that! We just launched a feature a few weeks ago to pay mana to "boost" your market in other users' feeds, which prods them to bet in your market and make the forecast more accurate.

In addition, we are planning a new feature that would go even further along these lines based on my hackathon project a couple weeks ago: https://manifold.markets/q-and-a

You can ask any question and post a mana bounty for the answer. This is the simplest mechanism for directly buying information. As a side-effect for getting the answer, you are also publishing it publicly for others to learn from. Just like public prediction markets, bountied questions are a public good!

All this ties into our vision of finding as many uses for mana as possible to build a strong information economy.

Austin has said that mana is the currency for attention. In an increasingly online and super-intelligent world, what could be more valuable?

Cheers!

Expand full comment

"There’s a lesson in this. I don’t know what the lesson is, but I am sure it exists."

The lesson is pretty obvious, indoor outdoor cats are the dominant lifeform on this planet, and when the AI generates superintelligent robot cat superbeings we will not be enslaved to them because we will yield ourselves in servitude to them voluntarily.

Expand full comment

Addicts often have to abstain and this can cause social difficulties. Say, avoiding alcohol in a culture where social interactions are centered around drinking -- this might even cause career difficulties.

But I think if the "prediction markets" dream comes to fruition, gambling addicts would face this kind of problem in an especially strong way.

Imagine the world where having a good public betting track record is a de facto requirement to get hired for a large number of desirable "normal" jobs. (Right now this is only true, maybe, in a tiny number of jobs in finance.) Or internal prediction markets ala Hanson take off, and now normal corporate middle managers are expected to be making bets all the time within their companies.

We are very far from this, but it is the dream of some proponents of prediction markets. In this world, if a gambling addict wants to completely abstain from placing bets, they'd be cut off from jobs that required placing bets, which might be large numbers of white-collar jobs. (And, if "futarchy" took off, completely cut off from participating in much of civic life).

I think "medium-large" success of prediction markets would be great and useful for society. But I hope they never become central to society, so that people can always opt out and have a completely normal life.

(Note: while the behavior described on Manifold at the beginning of the post definitely would seem to count as an instance of "problem gambling", I have no idea if there's a pattern of other behavior, so I do not want to imply that this person is a "gambling addict".)

Expand full comment

"But it was still a challenge the idea that it’s possible to run any kind of gambling-adjacent institution ethically, no matter how careful you try to be."

What exactly is unethical about giving adult fools an opportunity to part with their money, if there's no deceit involved? (Also, the quoted sentence misses "to".)

Expand full comment
(Banned)May 23, 2023·edited May 23, 2023

>>I don’t want to exaggerate how worried to be about this. People lose way more money on sports betting and poker every hour.

Yes, because sports betting is an international industry adjacent to some of the most popular, mainstream, and celebrated cultural and recreational pastimes, and poker is the most popular card game around. By comparison, prediction markets are the hobby horse of a relatively small handful of nerds who say things like "raise the sanity waterline" (No offense intended).

Isn't it possible that prediction markets are actually no better than sports betting or poker, except for in the ways in which their relatively miniscule size limits the damage they can cause? Isn't it probable that the damage would scale as (and if) prediction markets become more popular, up to the point where they cause as much harm as sports betting?

>A site that produces lots of great information,

Does it?

>raises the sanity waterline

Does it?

>and once a year or so causes someone to lose $29,000 which management immediately gives back because they feel bad

For someone who writes so extensively about slippery slopes and slur cascades, you sure seem very convinced that _this time_ this is as bad as things will ever be.

>is hardly the face of problem gambling in America.

In the same way that Fentanyl is the face of opioid addiction in America, and so heroin must be relatively okay?

--------

Do you remember when the FTX thing happened, and you were very distraught, Scott? You were caught completely off guard, who could have seen this coming, etc? This is one of those warning signs, happening before your eyes. Update your priors accordingly. Then again your solution to the FTX debacle was to lean even harder into the gambling markets, so I probably shouldn't hold my breath.

Expand full comment

Manifold conditional market on S&P drop/debt default is formulated as "will the S&P 500 at any point in 2023 drop at least 20% from its 2023 highs in 2023?" Since S&P500 is now close to year-high, conditionally on default, it triggers if the maximum drawdown of S&P over the remaining 7 months is more than 20%. But this event is pretty likely even in a world where the default has not impact on equity markets whatsoever. Back of an envelope and with current parameters, I would guess that the expected value of the drawdown is close to 15% with standard deviation of 5%, so a 20% drawdown is pretty likely (30%-40%?) without conditioning on any dramatic event. It is not 55% likely, so the market is saying that default would have some impact on S&P, but nothing dramatic, just equivalent to a return to 2022 levels of market volatility.

Expand full comment

> There’s a lesson in this. I don’t know what the lesson is, but I am sure it exists.

The lesson is "ladbrokes.com", which I am currently monitoring for the odds of the upcoming US presidential election. It's clear that the bulk of the utility (as measured by the humans who do it) of prediction markets is the thrill of gambling, not the intellectual curiosity of eggheads trying to get a better fix on the future of humanity. I think a prediction market could be built to serve the latter but not the former, but only by explicitly policing it that way, and there's not much evidence that it would attract enough participation to be useful -- after all, the first sort of utility accrues to the person participating in the market, and the second sort is externalized.

Expand full comment

Only here to say that my mantis ootheca just hatched and I am the proud father of 400-600 baby mantises

Expand full comment

I think that the Wale vs Minnows thing is not really gambling, but adjacent.

Traditional gambling is fueled by the possibility to make money: without stakes, roulette is a very dull game. This feels more like loot boxes or predatory mobile games, where you spend real money to get in-game benefits. (Of course, with such games, this is by design. From my understanding, the people who spend thousands a month are basically funding these games.)

In general, I also think wagering on real world outcomes is more of an addiction risk than wagering on pure games of chance: it takes a certain amount of stupid to convince oneself on can get an edge in roulette. By contrast, convincing oneself that one can beat a constrained market in e.g. election (or sport) predictions or something is not prima farcie absurd, IIRC our host claimed as much. It takes exceptional epistemic hygiene to account for overconfidence biases.

I would also challenge the assumption that there is no way to convert back "mana" to real money. Like life, the market finds a way. The obvious approach would be for two parties to arrange for the cash transfer offsite and transfer the play currency using highly specific bets ("Will the date between A and B go well" is not likely to attract any bets from the general public, and if neither party disputes the resolution, I do not see how Manifold could investigate the outcome (or even existence of A and B) independently.) Look for smallish accounts making money in big, cash-rich, tribal-coded markets (where the predictor has an edge) only to lose that money in niche markets with just a few participants.

I suspect that many zero-sum games can be implemented using prediction markets. For games of chance, just bet on the least significant bits of some stock index. For self-referential markets, weird stuff may happen. (I think one could implement a zero-sum variant of the dollar auction by asking "which of these options will trade highest at $date". I am unsure if the emulation capabilities extend if one allows the markets to refer to other markets as well.) Even if there is not a way to resolve a question fractionally ("answer a pays out 0.5$, answers b and c pay out 0.25$, answer d pays out nothing"), this could be emulated by using a random source to decide on any single outcome in the end.

Expand full comment

Dw, Medlock knows the moral of the story: https://twitter.com/jdcmedlock/status/1653552007385616385

Expand full comment

At time of writing the post about the S&P and defaulting has 7 traders. Scott, how much predictive value do you think a market like that has? I would think very little, and if so I think deserves a caveat or something if mentioned. But I’m very curious how you think about low volume predictions!

Expand full comment

"Congress is debating raising the debt ceiling. If they can’t compromise, the US will default on its debt, with potentially severe economic repercussions."

The use of the word "will" in your sentence is debatable. The Federal Government has not run out of cash. It receives more than $250 billion every month. https://fred.stlouisfed.org/series/MTSR133FMS

Annual net interest payments this FY are projected to be $640 billion. I.e. ~$55 billion per month. https://www.cbo.gov/publication/58946#_idTextAnchor004

The Federal government will not be forced to suspend interest payments on the debt. Principal payments are not a problem because they reduce the outstanding debt which increases the amount of borrowing that is permissible.

The FG would have to suspend or delay other payments such as employee salaries. But, that is not a default.

The FG should not be forced to miss payments of social security pensions because those involve reducing the outstanding debt (remember the trust fund?)

My argument in more detail: https://www.wsj.com/articles/the-phony-debt-ceiling-calamity-x-date-default-social-security-roll-over-treasury-ac939d83

Further, the FG has defaulted in the past. In 1933, the FG refused payment of bonds in the form of the gold coins that had been borrowed and were owed and paid only in depreciated paper dollars. SCOTUS declared the action unconstitutional and then did a somersault with a double twist to avoid payment of damages. https://www.law.cornell.edu/supremecourt/text/294/330

Do not ever believe the Federal Courts will protect you from the Federal Government when they decide to trample on your rights.

Expand full comment
May 23, 2023·edited May 23, 2023

"Team Minnow started cheating first. They rounded up their friends and asked them to register Manifold accounts and join the market. This might have been semi-fair to start, but then they started paying people, in real money, to do it."

How is this cheating? Does Manifold have an express rule about it and who can sign up? I tried logging in to read the original post about the Whales and Minnows and had to join, and Manifold were delighted to throw some of their play money at me to start a market of my own even though the only thing I wanted to do was read the article.

So it doesn't seem like they're any too fussy about who signs up, and if you're relying on "gentleman's agreements" or a sense of honour about not getting people on your side by paying them to be on your side, that boat has long since set sail. Honour is for the birds, the only thing that matters is winning (see all the debates about academic cheating and those willing to defend the proposition that it's okay, everyone is doing it, and the university is supposed to hand over the piece of paper in return for your money anyway so exams are just archaic nonsense, you can Google everything you need to know on the job).

Re: Sean McElwee, I got to this part and couldn't help but laugh:

"A bunch of guys sat around a table: a spokesman for Facebook, a head of an organization attempting to end the filibuster, a former top aide to former Senate Majority Leader Harry M. Reid, a senior reporter who covered the Senate for MSNBC, and Gabe Bankman‐Fried, the brother and political confidant of the crypto billionaire Sam Bankman‐Fried.

...Gabe Bankman-Fried, whose role involved helping his big brother figure out how to spend his money here in town, showed up semi-regularly to these poker nights. Gabe’s organization, Guarding Against Pandemics, was becoming a powerhouse in Washington, and Sean had been doing some work for them that involved hyping their work at every opportunity."

The Bankman-Fried name just keeps popping up, doesn't it?

And in light of the recent survey on here, this part also made me laugh:

"“All of the Zoomers that work for me are bisexual, and all of them have long covid,” he said. “I’ll believe long covid is real when someone who is not bisexual has it.” (This was a joke. He later told a fact checker he didn’t even know the sexuality of his staff.)"

I get the impression McElwee is a bit of a dick, but he does sound funny (if you're not too sensitive). But also do not follow his advice on gambling.

Expand full comment
May 23, 2023·edited May 23, 2023

"I cannot begin to predict the name of the US President in 2050"

Sure you can! You can't tell me it'll definitely be John, but I bet you can rank the following names by likelihood: Michael, Jessica, Sang-hoon, Stardust.

Expand full comment

> he predicted corona

> he then offered a 1m future on bitcoin

> wtf

I also was fairly early on corona and big into bitcoin; I think he's just bearish on the current civilization. I cant speak for why Id believe that bitcoin offer was a good idea, but for a model of bearish on civilization takes, imagine a dying man with several organ failures and several airborne illnesses and cancers, you don't know how he's going to die but the systems maintaining the status quo seem corrupted and destroyed.

I don't follow the news like I used to, but maybe something spooked him hard and when ask "will the dying man who just vomited blood die soon" he said yes.

Expand full comment

Two big bets failed in 2 predictable ways, wish you could bet on those bets at the time

Expand full comment

This article has two different people setting their money on fire to either actually (or subsequently rationalize as) promote awareness, Balaji's 1m about hyperinflation and Isaac's 4k about manifold. Balaji's 1m seems to have gotten him some press but little else, while Isaac's 4k got him some press and many manifold policy changes which are purportedly exactly what he wanted to begin with.

By that comparison I think Isaac did pretty well

Expand full comment

> he predicted corona

Well he gave a good estimation on what would happen if corona was to be a deadly pandemic, all of which was pretty easy to predict as far as I can see. Some of it was happening outside the US as he predicted it. Others were standard procedure.

Expand full comment
founding

re: the likelihood of the US govt being late on a payment, note that there is a betting market that I believe has $5-10bn bet on it, which is 1yr credit default swaps on the US govt. If a default is declared, the person betting on a default gets about 40% of the notional bet, just because there happens to be a $60 US Treasury bond (because it was issued when interest rates were a lot lower).

Right now that price of that is about 170 bps, so the implied probability of there being a declared default under this contract is about .017 / .4 = about 4%

But also note that there's likely (the contract isn't clear) a 3 business day grace period before a default is declared, and there's also the likelihood that a committee made up of mostly US banks won't want to declare an event of default if they get a call from Treasury saying they'd prefer not to see that. So e.g. if you think that there's a 75% chance that a delayed payment either is cured within 3 days, *or* for whatever reason the committee doesn't declare a default, then this is consistent with a ~15% chance of a late payment.

(btw the normal objection to "why would anyone bet on US govt default because there will be noone left to pay them if the US govt is bankrupt" I mostly don't think applies in the case of what everyone agrees is very unlikely to be a permanent default but rather "just" a delay)

Expand full comment

The markets on timeline of debt ceiling resolution and probability of default aren’t consistent. Consensus estimate for the X date put it prior to June 15 quarterly tax receipts--so to have only 80% chance of resolution by July 1 but 6% chance of default implies a 70% chance that the X date is actually into July (yes there could be a more subtle joint distribution of the two but still strikes me as off)

Expand full comment

"And at the time he made the bet, he didn’t really sound like someone trying to communicate that he only thought there was a 10% risk of a near-term crisis:"

A 10% chance of a severe crisis is not good odds at all. You should definitely be worrying about a severe crisis if there's only a 90% chance of it not happening.

Expand full comment

Defaulting on a bond is different from "defaulting" on payment for goods and services.

Defaulting on a bond means when the bond is due the government says "we're not going to pay you the full face value". It's pretty cut and dried. This is generally disastrous for the rate the government has to offer for any future bonds it issues. If a lot of the bonds are owned by a different government with an army this can be really bad, as in "We're going to invade and take stuff in kind for repayment."

"Defaulting" on goods and services usually means "we're going to stretch out the payment schedule." This is a process of negotiations. Businesses do this stuff all the time; almost everything is "net 30" or "net 90" or something to that effect. Not paying all the bills by the end of that "net" period is bad for the business's credit rating but usually doesn't result in the business going bankrupt. This would be unusual only because the American federal government hasn't done this in a long time (probably the Civil War was the previous time). But cities do it (see New York in the 70s, or for that matter I'm pretty sure that Detroit is having problems now) and states have I think done it as well.

Expand full comment
May 23, 2023·edited May 23, 2023

Is it possible to improve your own predictive accuracy by trying to guess the answer to “what would the average of a large number of people’s estimates of this be?” rather than “what do I estimate this to be?”?

Has anyone tried this?

Expand full comment

> But it was still a challenge the idea

should be "to the idea" I think :)

Expand full comment

Re: Book review contest; it doesn't look like Cities got a big boost? The currently-leading one in the linked market is The Educated Mind (unless I'm misreading).

Expand full comment

Re brinksmanship: I dislike watching Congress and Biden playing games of "chicken" with the debt ceiling, but, at least, unlike Putin, there are no nuclear warheads involved.

( One nit, partially covered in your post: If an agreement _isn't_ reached, there are a couple of never-before-tried alternatives that might evade the debt ceiling that have been discussed this time around: the trillion dollar coin, declaring the debt ceiling law unconstitutional, and I think one or two more. I don't know if they make the overall situation better or worse... )

Expand full comment
May 24, 2023·edited May 24, 2023

I don't think that tweeting "what if this is a pandemic" on Jan 30th is especially "prescient" when The Economist had a cover story on the same day saying "Will the Wuhan virus become a pandemic? Probably.".

https://www.economist.com/leaders/2020/01/30/will-the-wuhan-virus-become-a-pandemic

Expand full comment