The open source community work in ML tends to be terrible at the science and rigor parts of it. You see tons of lazy “we recreated an open version of X” that fails to be remotely as well done as the peer reviewed version, even from the larger community groups.