Korva is a scroll of the internet’s best ideas.
$100 million for one answer
The result is worth that much because the decision behind it is that big. The buyers would be giants like P&G, Coca-Cola, Nvidia, and Visa, with enough riding on one question to pay a nine-figure bill to a lab like Anthropic for the answer.
Picture Visa weighing whether to sponsor the World Cup for a 10-year exclusivity period. A model run that settles that one question is worth the price.
Governments and defense organizations are another likely buyer, since the same approach lets them speculate with data about the future.
How can a model simulate a market?
A common approach builds many software agents, each given the traits of a real group of people, and lets them react to a scenario. Researchers then count how the simulated population splits across choices. This agent-based method has been used in economics and epidemiology for decades. Large language models let each agent respond in words rather than by fixed rules.
Why do governments run simulated futures?
Military planners have run war games since the Prussian Kriegsspiel of the early 1800s. In the 1940s, the mathematicians Stanislaw Ulam and John von Neumann developed Monte Carlo methods on the Manhattan Project, which estimate outcomes by running a problem many times with random inputs. Defense groups later applied these ideas to forecasting and planning. The RAND Corporation, founded in 1948, ran scenario exercises for the US military through the Cold War.
Why do prediction markets forecast well?
Prediction markets let people bet on outcomes, so the trading price becomes a probability estimate. Traders who are wrong lose money, which pushes out poorly informed guesses. The Iowa Electronic Markets, run by the University of Iowa since 1988, have often matched or beaten polls in studies of their election forecasts.
More from 20VC with Harry Stebbings
