> the model is allowed to train on pretty much the test case knowing the ground truth The task is to solve the integral symbolically, though, right? It's a ha

> the model is allowed to train on pretty much the test case knowing the ground truth The task is to solve the integral symbolically, though, right? It's a hard problem to solve, even if the model is given access to a numerical integrator tool it can use on the main problem itself.