The poor readability bit is quite interesting to me. While the model does develop some kind of reasoning abilities, we have no idea what the model is doing to c

The poor readability bit is quite interesting to me. While the model does develop some kind of reasoning abilities, we have no idea what the model is doing to convince itself about the answer. These could be signs of non-verbal reasoning, like visualizing things and such. Who knows if the model hasn't invented genuinely novel things when solving the hardest questions? And could the model even come up with qualitatively different and "non human" reasoning processes? What would that even look like?