Generally because there's not that much to comment on, as this is effectively a somewhat prettier ungoogled chromium, and so there's not much room to make positive comments. Though I agree people get caught up in negativity easily.
It is some mix of priorities along the frontier, with Apple being on the significantly controlling end such that I wouldn't want to bother. Your trust should also be based on prediction, and giving a major company even more control over what your systems are allowed to do has been historically bad and only gets worse. Even if Apple is properly ethical now (I'm skeptical, I think they've found a decently sized niche and that most of their users wouldn't drop them even if they moved to significantly higher levels of telemetry, due to being a status good in part), there's little reason to give them that power in perpetuity. Removing that control when it is absued hasn't gone well in the past.
It certainly could, and I wouldn't be surprised if the authors want to try it out on those. You do have issues of past improvements often not quite enhancing more powerful models nearly as much. I'd expect this to possibly not work as well, something like the bigger models ending up with more polysemantic neurons because they're given more ''incentive'' (training time, neuron count, dataset size which they're encouraged to be able to reconstruct) to extract as much possible. This might make so the method performs worse due to this intermingling.
(See the transformer circuits website for that)
(Though I expect there's ways to recover a good chunk of extra lost throughput/accuracy, maybe by doing extra steps to directly steer the training towards breaking apart polysemantic neurons)