This has probably changed over time. My impression when I was taking college entrance tests, was that the ACT tested what you knew (i.e., memorized facts), but the SAT tested _how you think_ (i.e., how good you were at figuring things out). But that was in the early nineties, and they changed some things not very much later that, among other things, resulted in more students getting higher scores, which I think was the goal then too. They had a lengthy explanation about keeping the test relevant to the changing expectations of modern institutions of higher learning, but reading between the lines, it seemed like the main outcome was giving out higher scores.

Eh. Win11 with 8 GB of RAM might work if you never connect it to the internet, or find some other way to block Windows Updates from ever happening. (Maybe once it goes EOL, they will stop issuing updates?)

But every time it starts downloading Windows Updates, it's going to try to store half the internet in system memory, and the platform's horrible virtual memory system is going to consistently swap out the page that's going to be needed next, every single time, and the download that *should* take a few minutes, is going to take more than a week, during which time you can't use the computer for anything else because it'll be constantly swapping like it's 1996 all over again; and by the time the whole update is downloaded and installed, Microsoft will release another update. Rinse, repeat.

4GB of RAM, I can only assume, would be worse, if that's even possible.

16 GB is mostly usable for basic computer tasks like browsing the web. Mostly. But it's not great. I consider 32 GB to be the practical minimum, if you want anything resembling decent performance, on Windows 8/10/11.

Linux systems can run on 8 GB of RAM. Heck, depending on what you're doing, 1 GB of RAM will do ok, though that's not going to give you much of a desktop environment. It's fine for a lot of headless roles, though.

I don't think i386 builds would be usable on an ARM system anyway.

At worst, you can always just compile your own. Granted, the Mozilla codebase is (last I checked, which admittedly has been several years) significantly more of a pain to compile than the average open-source project, but it's not _prohibitively_ difficult. You do have to read a few lines of documentation and maybe edit a small config file, but there's nothing really tricky about it. Frankly, it's easier than installing most third-party binaries. Also, if you're using a distro that's made for your hardware, like Raspbian or whatever, it'll probably just have a package in the repo.

Yeah, I'm surprised and a bit disturbed that this hadn't happened long since. Linux distros pretty consistently started compiling everything for amd64 pretty much as soon as users had the hardware for it. There was no downside, because all of the software that everyone was using had source code readily available and could be compiled for the new system. It all has to be recompiled anyway, every time a major library (such as libc) gets any kind of really substantial update, because when you *have* the source code for everything, and everyone *knows* that you have the source code for everything, and your system includes a full working build chain out of the box, and compiling software from source is an extremely *normal* thing to do, to the point where people who aren't developers and wouldn't be able to read any of the source code don't have any trouble building it, it turns out that in that scenario hardly anybody bothers to maintain long-term ABI compatibility, because there's no compelling reason to do so. You (the package maintainer or whoever) can just do a fresh compile every time anything gets updated. You were almost certainly going to do so anyway. Even if the update only changes the documentation, you just build the thing, because it's significantly easier to just build the thing every time, than to bother to figure out whether you need to do so.

Certain other systems, that I don't need to name, took a decade or more (after folks had amd64 hardware) to transition over to widespread deployment a 64-bit OS and are *still* routinely using 32-bit applications in 2025; but that is for reasons that have never been relevant for Linux users. A lot of the users of such systems, would probably find the above paragraph just about as baffling, as the WWII Japanese naval commanders who found out the Americans had entire ships dedicated to making ice cream. "They can compile software so easily, that they do it when they don't even need to do it? They've already won the war, we just hadn't realized it yet."

A computer or robot with human-like intelligence isn't something anyone knows how to start designing. At all. We're no closer to that today, than we were in the seventies. No one even knows how to start doing research that would eventually lead to knowing how to make that. (No, LLMs are not heading in that direction. At all. Stop reading OpenAI press releases. LLMs generate output that is statistically similar to the training data. That's all they do. That's all they will ever do. It's cool and impressive when it works well, and it has some uses, but it's not even remotely similar to general-purpose intelligence, not even *vermin* level intelligence, much less human-level.)

Also, as technology in general and robotics in particular has advanced, we have consistently moved, over time, further and further away from the idea that a humanoid form factor makes for a particularly useful type of robot. Humanoid robots are usually created in quite small numbers (often, one-of-a-kind) and very consistently exist mainly to be shown off for publicity purposes. They almost never do anything really practical, and even when they do, they don't do it as well as other robots. Some of them (e.g., ASIMO) are subjectively really cool, granted. But they're not practical.

The human body is enormously practical, mostly for reasons a robotic system can't mimic at anything even vaguely in line with our current level of technology. Superficial things like the bipedal shape with arms and hands and fingers, mostly do what we need them to do, but they are not the main selling points of the design, compared to, say, a robot with treads and one arm.

> and because they don't want to reveal just how
> hot-garbage their underlying code is

Code quality isn't the issue. I have no idea what their code quality is, it might be fine, it might be terrible, but the reason I can't tell is because that has nothing at all to do with the problem with their results.

The fundamental problem is that they've been actively trying to convince a lot of people, up to and including their shareholders, that the product is a fundamentally different thing than what it actually is. They use fancy terminology that most people don't really understand, like "neural network", to actively disguise the fact that the product is, at its core, basically just running statistics and spitting out statistically-likely combinations of tokens. It's _basically_ a really heavily souped up Markov chain generator, on really powerful steroids. The most important steroid in question, is an absolutely stupendous quantity of training data. But there are also some more clever things going on, e.g. with the details of how the data are tokenized, and I think they're more clever about how combinations of tokens work, than just the flat one-dimensional sequence of a traditional Markov chain. All of these enhancements make the output feel much more similar to real human speech or writing or whatever, than was possible even ten years ago. But fundamentally, that's all the thing is doing: generating output that's statistically similar to the training data.

There are definitely some actual uses for this technology, but they're not even vaguely commensurate with some of the things the companies involved want you to *imagine* the technology can do; and they never will be, no matter how much the technology matures. LLMs are not a path to general-purpose AI, no matter how much people want them to be. We're not materially closer to knowing how to create general-purpose AI, than we were in the seventies. For that, we're still waiting on some fundamental and completely unpredictable breakthrough. This doesn't mean the technology is useless; it's not; it has uses. And as it continues to mature, it'll be even more useful, for the sorts of things it's useful for. But it's not going to make humans obsolete, or do a lot of the other preposterous things the hype machine predicts.

Yeah, short positions are very dangerous for an ordinary person, and probably best left to investors with very large and diverse portfolios, such as hedge funds and financial institutions.

Although, taking an indirect interest in a short position (e.g., by buying a few shares of a publicly-traded fund that happens to hold, among other investments, a short position on something) is noticeably less dangerous, as it tends to insulate you somewhat from the worst outcomes.

The *safe* thing to do, if you think a given asset is significantly overvalued ("in a bubble"), is to just not take any financial interest in it at all. That sounds boring, but in the context of things that can potentially have arbitrarily nasty repercussions, boring can be good.

Juries can be unpredictable, though, and enough money could potentially be at stake (eventually, assuming the technology matures and improves), that it's probably worth their while to have a go. Especially if they think other states might follow suit, because if they can manage to get a court to rule in their favor, that sets a legal precedent. Although, if they were really smart, they might want to be picky about which state they bring the suit in, for best odds of a favorable precedent; I haven't studied this question in any depth, but off the top of my head I'm guessing Texas *might* not be the most strategically sound choice. Possibly. But it's possible they may have considered this and determined that being proactive is more important than being strategic in the choice of state; if so, I haven't studied the situation well enough to gainsay that.

For the pedants, I probably should have clarified that you can, of course, use terms like "butter-flavored" on the packaging, pretty much no matter what your product is. Although, depending on the product, that may or may not convince anyone to buy it. "Butter-flavored" popcorn, could be good marketing. "Butter-flavored" popsicles, probably not so much.

These days, it might actually sell better if dyed blue. Neon blue, even.

I think you're still not allowed to print the word "butter" on the product labeling, if it's made from vegetable oils instead of dairy. Maybe there's an exception if you use a qualifier, e.g., "plant-based butter"? Not sure.

Not that generic words on the product label matter very much to most consumers. You're allowed to put a picture of the product on there, in context with other foods (e.g., spread all over a muffin), and your brand name, and marketing hype, and "gluten-free" as long as it doesn't have any gluten in it (even if it's something like butter that no reasonable person would expect to contain gluten), and so on and so forth. Lack of the word "butter" doesn't really stop anyone from buying it. When I was six or eight years old, and learning to cook (which almost everyone learned to do back then; unlike now when it's a hobby for only some people), I quickly learned that "margarine", "oleo", "butter", and "shortening" are all synonyms with exactly the same range of meaning in practice. Anyone who says otherwise has bought into the marketing. There _are_ differences between different brands (principally, in the ratio of water to fat to air bubbles), and depending on what you're doing with it those differences can be important; but which of the aforementioned generic words is on the package, is not a useful indicator of any of that. The FDA-mandated nutrition label contains more useful information.

I don't very often wish I had mod points, but this is an exceptional case.

Yes, but that won't happen until we get everyone on the same page moving forward in a fault-tolerant and robust expectations paradigm. In order to action that, we need to empower and enstrengthen key team players and integrate a results-forward meritocracy with our strategic core competencies; and if we could count on you to be on board, that would be great.

That's the official line, but it's a blatant lie.

To the extent that Google cares about account recovery at all, they want it to go away, and you get a new account every time you get a new phone. They've been systematically *removing* ways to recover accounts. Among other things, you can't do it with an automated computer-generated-voice call (that reads you a one-time code) any more, and for a while now you haven't been able to do it with just your password once they have your phone number: you have to have the phone (or at least the same phone number), or the account is dead, full stop. Additionally, they keep making it harder and harder to log into your account from multiple different devices, because they do NOT want you doing that. They want every account tied inexorably to a specific individual smartphone, preferably an Android phone (though they also work with Apple, reluctantly, because Apple users are ridiculously loyal and too numerous and too moneyed to completely ignore). For the time being, if you've given them an email address that goes to a third-party email account (from e.g. an ISP or employer), you can still get a one-time code sent to that account every time, and thus log into the Google account from wherever you are on the network; but I expect that option to go away before too many more months pass. Also, needing to do it kind of defeats the main purpose of GMail. You *should* be able to just log in with your password, but that's no longer allowed, unless you are on the same device you've used before. So if you're ever going to get a new phone, better do it before you lose the old one, or the Google account will die.

I am *guessing* that the motivation behind all this, may have something to do with smartphone apps, and the lucrative nature thereof; but this is a guess.

I know people who still *try* to use Google, and then they invariably come to me asking why it can't find what they're looking for any more, or why it gives the wrong answer now, or whatever. And I'm like "They decided to get out of the web search business, because they think there's going to be more money in AI." And the users just look at me confused, because they don't know what any of that means. "But why doesn't my Google work any more?" Eh, keep it simple: "Because Google is broken. It's not just you, it's broken for everyone."

What I wish I could give them, is another search engine that works anywhere near as well as Google did in 2021 or so. But there isn't one.

Or you could just, you know, fill out a 1040. It takes about ten minutes. A bit longer if you've never done it before.

People are irrationally afraid of it because there are so many horror stories out there about people spending hours and hours and hours trudging through financial records trying to figure out their taxes, but most of those stories are heavily exaggerated, and 100% of them are from people whose finances are *way* more complicated than average, because they own a business or have a bunch of fancy investments or whatever. For a regular person who has a regular job and gets a regular W-2 from your employer, it's really not a big deal. Though of course if most of what you know about it comes from the advertising from Intuit and H&R Block, you wouldn't know that.