Our first demo is a baseball at-bat.
v0 is a proof of concept. It plays out a pitcher against a batter, one pitch at a time, from the set to the swing.
The batter decides from what he can see before the ball arrives.
The model plays out that decision, and the pitch that led to it. Most sports models instead fit a line through past statistics.
We started with baseball because the game breaks into separate pitches, each one closely measured.
Each answer is a range, with what happened shown beside it.
So a price built on that range says how sure it is.
- The model throws the same pitch again and again and gives every place it could end up, next to where it went.
- The model estimates the batter’s read: what he could expect from what he had seen when he had to decide, next to what the ball did.
- A batter can choose well and still ground out, or swing at a bad pitch and get a hit. v0 gives the decision and the result as separate outputs.
Games, seasons and careers come next.
We’re now building games and seasons, then whole careers, which is what contracts are priced on.
Shown by invitation.
v0 is a research preview, not publicly released. We show it with its model card and technical report, and a technical whitepaper on how the models work.
Request access to v0: EMAIL_TBD