Or, if not purely synthetic, how about deliberately filtered in weird ways. Could you train a model where information about a specific scientific theory has been thoroughly censored from the training data? Then in testing you could give the information which was available to the human orginators of the now accepted scientific theory. See if the model can figure out the puzzle. Be concerned if it does better than you expect.
Or, if not purely synthetic, how about deliberately filtered in weird ways. Could you train a model where information about a specific scientific theory has been thoroughly censored from the training data? Then in testing you could give the information which was available to the human orginators of the now accepted scientific theory. See if the model can figure out the puzzle. Be concerned if it does better than you expect.