The entire premise of what that prize is is meaningless.

Basically an AGI should be able to exist as an algorithm initialized with random weights, and then learn everything from scratch. Training models on large data sets is not even remotely close to AGI.