armchair review here -- the paper is a lot of work by a large team.. the math is very challenging, and appears to be thorough.. the field is fast-moving so the

armchair review here -- the paper is a lot of work by a large team.. the math is very challenging, and appears to be thorough.. the field is fast-moving so the paper is useful as a "list of other current papers" if nothing else; gets the hard-working team some recognition, and spreads the sophisticated maths among more practitioners at a time when this is internationally significant.. My guess as to the importance here is that the "mesa" technique might "steer meaning linkage" somehow, in the middle of "poorly understood" behavior. As another YNews other comment mentions, it may be more labor for unclear returns, yet it is based in theory that might lead somewhere over time. This paper focuses on NLP language once again, but the transformer tech is being applied to a LOT of digital domains. So this "mesa" theory might find application elsewhere soon. I do not know why this particular paper was upvoted (feedback welcome). A review of "useless" seems premature to me, and ignores other positive attributes, some of which are mentioned above.