Again, I don't know if this is something that is only true in Biological science and less so in Computer Science, but while I agree most research is a minor incremental improvement, its often on a topic so specialized that most outside of your research group aren't very familiar with the intricacies.
This begs the question, like you said, then why would you care about hearing that talk? Maybe you don't even care about their thesis and are just interested in adapting a technique they used in your own work. Or maybe their thesis bridges a gap between two fields of study that were previously not very closely linked. Besides, when you're trying to communicate your work to other people, isn't it to your advantage to try and cast as wide of a net as possible? Isn't part of the point of talks and conferences to widen people's spheres a bit? There's no need to leave your home lab if you want to converse with people who understand most of what you're researching already.
Say, I'm computer vision researcher specializing in (non-deep learning) 3D reconstruction. Why would I be listening to talks where someone goes into some esoteric tweaks they made to the deep learning stuff? I feel like there almost should be separate conferences for people doing feature recognition/tracking, 3d reconstruction, object recognition etc., as all those fields are vast in themselves. And, once you're on such conference, it's just the experts there and you don't need the 5 minute intros any more.