Many people have already stated that they are writing primarily for LLMs. Now, everyone can write for both LLMs and prediction markets. Humans are particularly bad at intuiting probabilities, and the institutions we’ve built reward everything except accuracy. We consistently overweight confidence, popularity, and aesthetics over ground truth, leading to a world where the loudest voices are often the least accurate. This is one of the root causes of the recent vibe shift. Changing what we value requires changing what we measure. This is why we built claim markets: the global leaderboard for all public claims. Claim markets translate what a person says across the internet into prediction market positions. We’ve kicked things off by tracking 63 people across 7 domains, and anyone can add additional people to track. There has been very little innovation in prediction markets, particularly in novel primitives and form factors. All solutions thus far are active, requiring users to submit trades and seek out information. Claim markets are the first passive solution, embedding context from anywhere on the internet into a prediction market position, and unifying it into a global leaderboard. Claim markets unlock questions that were previously impossible to answer. With sufficient data, they become an engine for discovering who has frontier context and analysis in a given domain. Some questions and areas of research I’m excited to explore include: - Expanding tracking beyond individuals to news outlets, blogs, and institutions. How predictive is Citrini, and which underrated, undiscovered blogs are consistent sources of signal? - With a sufficient number of tracked figures and data, we can start to answer questions like "when Nate Silver and Tyler Cowen disagree on US economic policy, who's right more often?" or "which combination of pundits gives the best signal on tech policy?" - Which public figure’s claims actually move markets? Prediction markets are an extraordinarily powerful tool to elicit information through finance. When utilized correctly, prediction markets create positive-sum outcomes by redefining incentives and enabling better coordination.

I'll be at Manifest in Berkeley starting today and will be giving a talk tomorrow at 1 pm. Hit me up if you want to chat about all things predictions or markets!

Market makers deliberately take on directional risk based on internal fair value models and expected flow. They are rarely completely outcome-neutral and are in the business of taking positive EV bets with variance.

Prediction markets teach you more about human behavior, incentives, and markets than any economics textbook. The best traders all exhibit extremely high initiative empathy to understand their counterparties. Most importantly, it prevents you from buying into EMH propaganda.

Polymarket orderbooks update ~2-10 ms faster than trades.

There is little overlap between top traders on play money markets (Metaculus, Manifold) and real money prediction markets (Polymarket, Kalshi). They are games with different incentives that attract different crowds and ways of reasoning about the world.

Agent-based research turns research into an optimization problem. Point enough resources at a goal, and the system gets very good at searching the solution space. It already feels like we're stuck in the econ PhD local equilibrium: more and more talent competing over legible, in-scope questions, while the bigger ideas look to weird to build or fund. Cue the 15th paper on the wage return to an extra year of school in rural Wisconsin.

Polymarket's current oracle design resides in a local equilibrium that is unlikely to change in the near future. It's an optimal setup for the main constituents (Polymarket, UMA, regular traders) for the two things that matter: liability allocation and trader retention.

Economic alignment is the design of the games and institutions around AI agents. Once agents act economically, alignment is not only about what a model does. It is about what the surrounding system can measure, reward, verify, remember, and route. Essay: cgi.md/essays/economic-align…

Hard to believe it's only taken a year to go from ultrathink to ultracode.

On Polymarket, sports politics pull from the same wallets, while crypto politics barely overlap

Estimated new philanthropic funding just from OpenAI Anthropic: $100B Entire DeFi TVL: $83B

Prediction market oracles are not designed to seek ground truth. They are designed around liability allocation and trader retention. The three actors (which may not be distinct sets) are: 1. The prediction market platform 2. The third-party oracle 3. Traders - Platforms (especially decentralized platforms) want credible resolution without directly managing every edge case. - Third-party oracles like UMA provide legitimacy, process, and a blame layer. For Polymarket, this abstracts away resolution liability and public blame while serving the users who matter most: the < 1,000 regular traders doing > 50% of Polymarket volume. - Core traders seek sources of edge, which in the LLM age is frontier information that moves through people, private channels, or undocumented sources before becoming legible to the internet. Today's oracles are designed to provide sufficient value to all three constituencies. The result is a stable equilibrium where no single party is incentivized to change unless an external force disrupts the equilibrium.

Top 1,000 addresses* The number of unique users is significantly fewer. Discretionary traders on prediction markets generate all the valuable information: the prices that millions of passive observers consume. Tomorrow's dominant prediction market is < 1,000 users away.

Prediction markets haven't hit PMF until their datasets have minted at least 1,000 PhDs.

The majority of users don’t care about platform risk, on-chain auditability, on-chain matching, or anything else that does not improve their product experience today. If you’re a builder that cares about these characteristics, you must work creatively around these constraints.

The path dependency on how prediction markets develop over the next two years will either set the industry back years (again) or evolve into the core infrastructure for understanding the future. The dominant venues today have the most immediate impact but this will change in the near future. While the current market structure is completely broken for prediction markets, the most important field going forward is creating novel products with clear positive externalities for users.

Lots of liquidity on the Polymarket orderbooks these days.