Do bigger AI models get better at simulating particle collisions?

Physicists tested whether the scaling laws that govern large language models also apply to neural networks trained to simulate particle jets from colliders. Model size follows the expected logarithmic scaling—bigger networks perform better—but dataset size and compute show weaker returns, saturating much faster than in language models. This ceiling likely stems from the inherent randomness of quantum physics: a network can only learn so much from finite training data before additional examples add noise rather than signal.