Completely agree with you there that the numbers are certainly correct. I just felt like they might not tell the whole story. For example, looking at the section where they say, I paraphrase, "x preprints are uploaded every day that mention TF", I don't doubt the numbers, but the way they tell it certainly evokes a different image compared to saying "the share of preprints relying on TF has been steadily declining the past years".

Regarding the many TF users out there, I would be curious what the main benefit is for them. Is it TPU support, TF serving, TF lite, something else?