I think that's just a limitation on the size of the model. I'm pretty sure that they use a pretty small model in those summaries to save money, which naturally makes them a little less smart.