Prediction Markets 2.0: Betting on LLMs
Sam Lessin
Context: I’ve been thinking a lot about the evolution of prediction markets and how they currently operate. In traditional financial markets, most trading volume is dominated by high-frequency algorithms snatching up fractions of a penny repeatedly. But when it comes to prediction markets — especially in crypto and related spaces — things are still pretty manual, slow, and limited. Mostly basement-level folks betting on small, discrete events like tennis matches. That just doesn’t scale.
Market Signal: The future of prediction markets has to look more like high-frequency trading, where machines compete by spotting tiny mispricings and trading rapidly. This isn’t just about AI for the sake of AI — it’s a natural progression. However, the biggest challenge is finding a scalable and reliable data source for these machines to trade on. Real-world oracles are slow and inefficient, so the digital realm demands a different approach.
Takeaways:
I believe the real breakthrough will come from prediction markets betting on the outputs of LLMs themselves. Imagine machines trading on what a specific LLM will output in response to a particular prompt at a future date — “On Dec 31, 2027, ChatGPT-5 will respond like this to this exact question.” This creates a clean, scalable, high-frequency data surface perfect for algorithmic trading.
This concept — which I call “Hayek’s Revenge” — highlights how markets are ultimately information processors. Betting on real-world events at high frequency is just too slow and messy, but betting on model-generated outputs is massively scalable and also has practical benefits. Trading in this space could help us better understand where models are accurate, stable, or failing, which is invaluable as reality itself feels more uncertain.
Ask: If you’re working on anything like this — high-frequency prediction markets based on LLM outputs — I want to hear from you. Traditional, human-driven oracle markets are fine, but I’m excited about machine-driven, high-frequency prediction markets with scalable data surfaces. Let’s connect!