What I feel curious about this month
There is a bit of a gold rush in prediction markets right now, spurred by the success of Poolymarket and Kalshi. For instance, here is $150M going into a prediction market fund. Wild!. In this rush, there is an opportunity to create an oracle resolution service. Does any aspiring entrepreneur want to do this?
Legal drama: There continues to be legal drama around Kalshi and Polymarket in the US. But it’s just like a never-ending crescendo, there is no definitive solution yet. Because circuits are split, this will ¿probably have to be decided by the US Supreme Court? Meanwhile Kalshi keeps printing money. Anyways, the new CFTC commissioner seems pretty sympathetic to crypto and prediction markets.
Automated forecaster startups scared me a little bit a few months ago, but they haven’t really launched yet, and I’m still getting various requests to do misc forecasting. When will the other shoe drop? I’m also torn between seeing how they would replace some of my skillset, while also being a really good tool I want to have at my disposal for Sentinel.
There is a Request for Proposals from Open Philanthropy on AI for Forecasting and Sound Reasoning. They mention “AI forecasters are currently nearing performance parity with pooled non-expert human forecasts, and we anticipate they will soon match or outperform top superforecasters”. Anyways, they are offering $8-10M in total funding for teams working in this space.
Prediction markets, forecasting platforms, teams
Polymarket will be integrated with Google.
With its massive growth, Polymarket is investing $1M in a builders program, increasing its market value by investing in its ecosystem. Seems like an interesting and more sustainable model, financing public goods because of selfish interest in a way that lets you do more of it in the future.
And a paper estimates that ~25% of volume in Polymarket is wash trading, i.e., providingt the service to Polymarket to juice their usage stats for VCs.
The NYC stock exchange parent company invested $2B in Polymarket
Kalshi valued at $5 Billion, and expands overseas
An updates on MetaDAO suggests there is nothing particular about a 50% pass/fail threshol, and suggests having a lower threshold (45%) for accepting proposals by the team, thus still providing some market oversight.
“We haven’t seen any impact (from sports prediction markets event contracts),” Caesars officials said”
Truth Social Teams Up With Crypto.com To Launch New Prediction Market. lmao. The Trump family now appoints the CFTC commissioner, but also has investments in Polymarket and Kalshi in addition to in this new venture.
The SuperMarket is a new platform. Seems like their profit model is selling the aggregate forecast. And here is a list of CFTC-regulated prediction markets.
We at Sentinel continued tracking large-scale risks, and interviewed Paul Scharre on drones and how AI will transform military command and control.
The Swift Institute is offering a paid subscription service at $10/month, focusing on the Middle East and North Africa. I hope them the best but at this price point the economics are brutal: you need 100 people paying you 1% of their income to support 1FTE, but getting 100 people to pay you is hard.
Research
Q2 AI Benchmark Results: Pros Maintain Clear Lead. Pro forecasters are better than engineer-led effots to replace them with LLMs. But months have passed since then.
Remarks on Bayesian studies from 1963.
A Handbook of Technological Forecasting and Roadmapping, by some Serbs.
This leaderboard tracks how good LLMs are at trading money. In some sense maybe an unsaturateable benchmark.
Odds and Ends
Manifest DC happened.
A programmer reports being able to better predict earthquakes. Curious about whether this pans out!
“Launched by the UN Secretary-General, the Early Warnings for All (EW4All) initiative sets an ambitious goal: by 2027, every person on Earth will be protected by life-saving, multi-hazard early warning systems.” If only.
People.ai has an AI-native forecasting solution which seems indistinguishable from every other product out there.
Weather forecasts on the timescale of 2 weeks to 2 months, are hard. But relevant to whether greater Tehran (18M people) will have to evacuate!
“We find that experts assign a median 5 percent probability to a large-scale nuclear event by 2045, while superforecasters estimate 1 percent”, says a Harvard study
DraftKings CEO: Bettors Will Choose Sportsbooks Over Predictions. Also DraftKings: DraftKings secures prediction market entry with Railbird acquisition; DraftKings Will Use Polymarket’s Infrastructure for Its New Prediction Market
Forecasting critical economic & political events via electricity consumption patterns in the United States of America and Turkey.
This takeoff tracker has a peculiar conception of how to present timeline forecasts, but is taking its headline data from this more serious Metaculus question.
Introducing Chronos-2: From univariate to universal forecasting. Ominous! There is also a similar TimeGPT, a foundation model for time series forecasting.
A prediction markets directory.
Articles
Friend of the newsletter Walter Frick writes What Would an AI Crash Look Like?
Some headlines:
Britain can defy gloomy economic forecasts, says Rachel Reeves. German economy ministry revises 2025 growth forecast up slightly to 0.2%, source says. Bleak
NYC Sports Gambling Case Shows How Mafia Clans Profit From Teaming Up
How quirky prediction markets evolved into sports gambling behemoths
Prediction Markets Are The Tax-Advantaged Way To Gamble On Sports.
A few of our clients who have studied the responses of different strata in their organization say that upper management who participate in their prediction markets have been proven to have a more optimistic forecast than what reality dictated. And those in the lower levels of the organization tended to have a more pessimistic outlook than reality dictated.
—Seminar Announcement: Using Prediction Markets in Government. Inkling Markets; Tuesday, August 31, 2010.
Clarification:
> The fund will consider a wide range of consumer-facing opportunities but will concentrate predominantly on online gaming, prediction markets, sports media, ecommerce, health and wellness, and interactive entertainment.
So not only prediction markets
Yeah, resolution oracle services was the only plausible profit center I could identify when I thought about how you would monetize a play-money prediction market / crowdsourced forecasting oracle. You could build a "Chinese wall" internally and have a different team of people pre-approving and resolving the questions, through an anonymization filter, not necessarily even knowing which are real live questions and which are test data. Once you have spun up this system and validated it on your own internal questions you could start to think about contracting it out. I wasn't sure if this was actually enough to justify the whole thing as a business but it was what I could find for a revenue generator.
The alternatives seem to be gamification targeting broke college students who are too poor for the real-money prediction markets (Manifold), or else some kind of vaguely defined forecasting consulting service like Metaculus is attempting that obviously isn't serious or likely to work.