No, it isn't. They call out ExIt as an inspiration as well as AlphaZero, and the implementation of these things (available in many of their authors' papers) is almost indistinguishable from LLMs. The architecture isn't novel, which is why this paper is about the pipeline instead of about any of the actual processing tools. Getting prickly about meaningless terminology differences is definitely your right, but for anyone who isn't trying to define a policy algorithm for a transformer network, the difference is immaterial to understanding the computation involved.