How would the rise of dash usage in LLMs have arised if a significant portion of non-LLM writers weren't inclined to take them up and make them more common? The only explanation I see is that they are common in training materials we don't as commonly consume as website visitors.

I have often wondered this myself, especially because the same stylistic quirks are found across models from different labs.

I haven't found a satisfactory explanation, but whatever the explanation is, it is undoubtedly true that LLMs use them to an almost absurd extent compared to the vast majority of human writers. Anyone who reads a lot of prose can see that.