I think it is more so just very common data in any dataset used in training. And thus you would expect model to produce it when relevant tokens appear. These m

I think it is more so just very common data in any dataset used in training. And thus you would expect model to produce it when relevant tokens appear. These models store these common patterns and then repeat slight variations of them as that is natural flow of the tokens. Humans do it similarly with the same problem if they have heard it before. Challenge is that coming up with actually novel reasoning problems is not that easy. So generating one to test models is non-trivial task.