Would two AI scientists disagree with each other, even if trained on the same data?
After seeing classical physics, AI scientists disagree at first but converge to known theories (Lagrangian/Hamiltonian) when data become diverse.
Check out our paper: arxiv.org/abs/2504.02822
