[godelski]

I hate Apple products, but I just hate Windows more and am willing to give Linux a pass more.

  Why are core util programs different?

Seriously! I know it's BSD based, but let's be real, you _should_ be able to use a command one-to-one from linux to OSX. That should be the goal. These are __core utils__! And you're gonna tell me that flags are different?

  Brittle AI features with obviously insufficient testing

I'll be fair, Apple isn't alone in this. But have you all heard about augmentations? Do you all dogfood? Seriously, I'll consult because when my partner is on the train I don't want her voice suppressed and the train sounds amplified, I want the opposite. This is a solvable issue... But your features are doing the opposite of what they are intended to. To all AI production people: augment the shit out of your data, scrutinize the shit out of your data, don't throw it at the wall and see what sticks.

Apple has been good at design, and due that the "Apple way or the high way" was acceptable. Because at least the "Apple way" made some sense. But now the Apple way isn't about making the product better, it is about making it thinner. I know that's easier, but that's the cost at being on top. Don't abandon what got you there.

[mananaysiempre]

The flags are different because GNU as originally conceived is the Borg of Unix userspace—every option and feature across 1980s commercial Unices must be assimilated (perhaps after minimal rationalization). Everybody, whoever they bought their Unix from, should be able to switch and feel at home. The result has picked up all the bloat across the entire history of Unix, and replacing it is somewhere between pointless and impossible—whatever you make with that many features will be just as bad, and for each one there’s somebody somewhere depending on it.

[minkles]

This could be more accurately phrased as Linux being an incompatible fork of Unix and macOS actually being Unix. And you are annoyed at this scenario.

I'm not sure what your noise cancellation point is, because that works absolutely fine here between myself and everyone else I know. As for the rest of the AI stuff, I don't use it so I can't comment.

Edit: on the first point, I'm not really a fan of some of the GNU stuff. I'd rather use FreeBSD than Linux for example. That is more logically consistent.

[godelski]

> This could be more accurately phrased as Linux

I do agree and acknowledge the point. But yes, I do find it frustrating that there are such differences in core utils. I guess the frustration is more about not working harder to find consistency. I many of the man pages it is not uncommon to find flags and options that exist simply for historical reasons. But many of the frustrating points are addressable in similar ways. An example might be with `du`. The OSX implementation has no `--max-depth` flag, despite `-d` being supported. Unless I am missing something, there is no downside to incorporating the long version of the flag. I can understand some other choices, but even `grep` is different enough.

> I'm not sure what your noise cancellation point is, because that works absolutely fine here between myself and everyone else I know.

In part, this is my point. Finding where generalization fails is non-obvious and non-homogeneous. So when it fails, it typically fails in a sub class of problems. These are of far more concern than edge cases[0]. Generalization is VERY hard. It is why dogfooding is so important as well as listening to the userbase. Because it helps you identify where you're lacking.

And maybe I caused confusion by my choice of words, thinking it would be simpler and context would clarify. I specifically mean sound isolation when speaking. As in she can ride the train, call me, and hear me perfectly fine (that side is working as expected). But what sound is being sent to me is not her voice (this is clearly suppressed as it has that same format as what background noise often sounds like), but the train itself. And worse, the train is amplified. So there is a identification in what is noise and what is speech. It may help to specify that this is happening when she is not in America. There are no issues with Bart, and that should tell us quite a lot. (I'll also add that augmentation is one of my niches as a researcher)

[0] The common approach to this is by using more data. Crude, but effective. Though it can exacerbate your issues if you don't expand the distribution you sample from, since you'll just reinforce your local density.