The rcond param is passed to np.linalg.lstsq, so the line becomes
scores -= p * (exposures @ (np.linalg.lstsq(exposures, scores, rcond=None)[0]))
The difference between the new and old behaviour is explained in the np.linalg.lstsq docs. I trust that the numpy overlords know what they’re doing, so I’m personally using the new default behaviour.
Hope this helps!