It’s not impossible to analyze and test compiled binaries.

When they say such things, the are probably talking about the expected value, where those chances are taken into account, just like the number calculated in this article.

That seems to be the opposite of what the others are saying: https://en.wikipedia.org/wiki/Autorun.inf#Inf_handling

Windows 7, Windows 8, Windows 8.1, Windows 10

For all drive types, except DRIVE_CDROM, the only keys available in the [autorun] section are label and icon. Any other keys in this section will be ignored. Thus only CD and DVD media types can specify an AutoRun task or affect double-click and right-click behaviour.[9][10]

You are remembering that the executable features of autorun.inf is disabled, which is still true. Autoplay (if enabled) as it exists currently only applies for discovered media file types and makes your default configured media player responsible for handling them. It would not be possible to execute arbitrary tasks unless you had an ACE exploit for the installed media player.

Sure, if they were designed that way, I would not call them defects either.

Just because all defect stock are routed to the US inventory, doesn’t mean that US inventory is made up of all defect stock.

If something happens I’ll make a switch.

To what?

AFAIK there is no need to re-encode, since Youtube videos are stored and served in chunks anyways. The change is that they are now slipping in the ad chunks as if they were a part of the normal video chunk stream.

Here’s an image viewer example with 0 exposed HTML elements (all UI rendered through a single canvas) and 0 human readable code (all client side code compiled to webassembly bytecode). Trying to block unwanted content in this kind of site would be closer to cracking a video game or patching an android app.

slowly open it up to the point where you can actually install regular linux apps on it

The linux running Chrome OS is completely separated, by design, from the virtual machine that runs linux apps under Chrome OS.

That article (or rather, the article linked in that article) doesn’t contradict your intuition, just a specific interpretation of that intuition. The randomly generated data puts everyone around 50%, which is indeed what you would expect from randomly uniformly generated data. So the similarity that the generated data presents is supposed to imply the conclusion that “everyone thinks they’re about average, so their judgement is no better than randomly guessing (assuming that the guesses are uniformly distributed)”, which is a subtle difference from “dumb people think they’re smart” - the latter attributes some sort of “flawed reasoning” to one’s self-judgement, while the former specifically asserts that there is absolutely no relevant self-judgement going on.

edit: You would also be correct that this doesn’t disprove the previous explanation, it just offers an alternative explanation for the observed effect. The fact that data matches up with a generated model definitely does not prove that it is not actually caused by something else, which is one of the criticisms of that viewpoint. It is obviously easier to rigorously demonstrate a statistical explanation than a psychological explanation of course, due to the nature of the two different fields.

Definitely not to have android apps on a Linux tablet, because in-waydroid rotation doesn’t work, and rotating the tablet itself breaks the windowing system until you reboot the container. Issue first reported in 2021.