In GameOut, you create forecasts by chatting with an AI agent. Its job is to define the scope and question of the forecast job with surgical precision. As you chat, it proposes structured "Forecast Cards" that you can refine through conversation. Once you're satisfied, click "Game it out" to launch the forecast. That's it. Your participation ends there. The rest is automated. About 1-3 hours and 3-4 million tokens later, your forecast is delivered. Check out our public predictions: gameout.ai/our-forecasts

Can I try it? Our public forecasting tool, that you can use on any question, is coming soon! To get early access, create an account at gameout.ai to join the wait list. In the meanwhile, you'll be able to subscribe to updates on our public forecasts. And, when we launch public forecasting, you'll be among the first with access.

Forecasts are ever-changing One of the things that professional forecasters do is regularly update their forecasts when new information becomes available. GameOut does the same thing. Periodically (every 4 hours, daily, weekly, depending on the forecast) GameOut checks for new information, and adjusts the odds. This loop is part of what makes GameOut so interesting. Each forecast is a living prediction. It can and will change over time. As decision-time approaches, GameOut checks more frequently for new information. When you are on the wait list, you can subscribe to any of our public forecasts, and get email updates whenever they are refreshed. gameout.ai

Is it effective? Maybe. We think so. But, we have to prove it. To do that, we have run a bunch of forecasts which we have published in public. We will run more in the coming days. Our initial forecasts will focus on near-term/intermediate events, primarily (think elections, court decisions, FDA approvals, corporate events like IPOs). Every call goes up before the result is known. Then we'll post what happened, including the ones GameOut gets wrong. gameout.ai/our-forecasts

How does it work? GameOut uses 25 different AI models in a custom orchestration harness to systemically research potential outcomes, the paths that lead to them, their likelihood and then to estimate probabilities independently. Once it's done, it loops and challenges its own assumptions, rethinks it, and applies a mathematical scoring system. It's a very similar process to what is used by governments and corporations, and their consultants. Think of it as a multi-LLM harness focused on tabulation, decomposition, game theory and Fermi estimates. The process is costly and uses a ton of compute, with most forecasts running for about 3 hours and burning about 4 million tokens. Explore the process here: gameout.ai/how-it-works

What does it do? It predicts future outcome probabilities using sophisticated professional forecasting methods and a ton of AI tokens. It starts by refining the question, then gathering the data and evidence, measure historical base rates, tabulating statistics, identify the players and their incentives, goals and salience, creating models and decision trees, and then digging into competing scenarios. It identifies questions, breaks down big unknowns into smaller, testable presuppositions, and then looks for evidence that contradicts them. The idea is to disprove, not to prove. The process is rooted in the principles of game theory equilibriums, data, Fermi decomposition, balanced inside/outside views, probabilistic thinking, Bayesian updating, scoring, and other actual, professional forecasting techniques. Only then do we put probabilities out. Explore our forecasting methods here: gameout.ai/methods

Why did we build it? Governments, hedge funds and big companies spend fortunes on super forecasters, consultants, models, etc. to build accurate forecasts of the future, just to stay one step ahead of the rest of us. The rest of us, normal people, can't do the same and are stuck guessing or asking ChatGPT to guess for us. We didn't think that asymmetry was fair.

Introducing GameOut: an experimental AI system that predicts the future It works by using a advanced forecasting methods (stats, fermi, game theory, war games, etc.) to systematically game out real-world outcomes (hence the name!) We're going to find out how good it is. In public. 🧵 ⤵️