When Will We Reach AGI—And Other Prediction Market Bets

On Friday, I wrote about how companies are scrambling to prevent employees from using confidential company information to bet on prediction markets, sites like Polymarket and Kalshi that offer derivatives to wager on events. Opportunities to bet on AI outcomes, say a model release, have exploded in popularity—bets for the latest OpenAI model reached nearly $10 million last week, for instance.

The bets may hold clues into what people who are—in theory—following AI closely expect about some important questions, such as the timing of artificial general intelligence, an AI system as smart as a person. For instance, OpenAI co-founder Ilya Sutskever recently said he expects an AI that can learn new knowledge as well as a person can will be developed in 5 to 20 years.

Kalshi and Polymarket, the two biggest prediction markets, give OpenAI an 11% to 15% chance of announcing that it’s developed AGI next year. Traders on Manifold, which uses play money, think it’s a coin flip whether AGI will arise before or after 2036. Two separate Manifold markets have Google DeepMind as the favorite to develop AGI first.