I would say "story mode" in papers is like dancing about your doctorate: really cool when the combination works, but the latter is where all the value lies, so

I would say "story mode" in papers is like dancing about your doctorate: really cool when the combination works, but the latter is where all the value lies, so it shouldn't sacrifice anything for the former. What about LLM C, which takes a set of papers as vertices, forms an abstract simplex of all their combinations, and then spits out new papers on the ten most interesting higher-dimensional faces?