Microsoft issues the secure boot keys that are used by all Linux distributions.

If they can just arbitrarily yank someone's keys like this, apparently without explanation or appeal, then what does that mean for those Linux keys? Are they subject to withdrawal for no reason as well?

Incorrect. Microsoft signs the boot shim. This lets you use Secure Boot with the default Microsoft keys you use to boot Windows. So any PC, with Secure Boot enabled, can boot Linux. The keys built into every PC are Microsoft's, and even if you hard reset the machine, they will revert to those Microsoft keys.

You are encouraged though if you run Linux, to create your own keys, and install them on your PC. Doing so would require you to re-sign the Microsoft bootloader but you are free to use your own keys. The only reason Microsoft signed the shim is because some OEMs do not make it easy to install a third-party key to secure-boot a non-Windows OS. So the Microsoft signed shim means if it can boot Windows, it can boot Linux.

And I say shim because that's the actual component signed - major Linux distributions re-distributed the signed binary. But it's bootloader independent - you can use the signed shim to boot your own version of GRUB or other bootloader and continue the secure boot chain if desired. (If you use something like Ubuntu, you're likely to encounter this if you try to compile your own kernel or module where you then h ave to add a key to the shim so the kernel can run your new module.

Microsoft can stop signing new shims, but that has nothing to do with Secure Boot. It's just a way so everything that can boot Windows can boot other OSes even if the OEMs lock down the computer.

Big companies often use their own keys for secure boot.

Then again, Apple's treatment of the Vision Pro perplexes me, so who knows what Apple is doing there...

I think Apple is trying to figure it out too, so they're just letting it be a developer's playground as in "Here's some cool hardware, now do something with it".

It's a device looking for a purpose, and Apple is just trying to see where that goes. I'm sure most of what we do with smartphones today wasn't what Jobs envisioned back for the original iPhone, so the Vision Pro is similar. Maybe hoping to see if an interesting use case pops out.

The pilots carry a transponder. If you make it down to the ground alive, you hide, turn on the transponder and wait for a rescue.

There was an article about the Boeing device that is just this.

It uses spread-spectrum wideband technology in fast burst mode so you can communicate. It's encrypted, digital and the fast burst means every transmission lasts well under 1ms which means by the time you detect it, it's too late to triangulate the position. (Think of the annoying beeping prank toys you can get)

Wideband technology means even if you pick something up, the location is spread out, and is a much weaker signal as it barely disturbs the noise floor.

It's also a 2 way communicator so you can send messages to it and they can reply back, and basically handheld. You strap it on, turn it on and it transmits your position in those bursts.

Of course, the biggest farce is about the second pilot being reported in the news. Which is a farce because if you know anything about the F-15E, you know it's a 2 man aircraft. So you know if you shot one down, there are two people you have to account for. It's not just public information, it's basically well known information. The fact that one was rescued means obviously there's one more to look for.

To put it in perspective, by then you could get IBM PC clones for under $1,000.

Not in 1984. Clones were cheaper, but the $1000 PC era didn't arrive until the 90s. The cheapest IBM PC was around $2000, but you were looking at like 128K of RAM and NO floppy drive. (The original IBM PC had a tape interface so you could use ... audio cassettes. While not unusual since every other computer had tape interfaces, basically nothing used it other than CBASIC - cassette BASIC which was a ROM BIOS option). That configuration was rapidly dropped and prices rose because you really needed 256K of RAM and 2 floppy drives.

Though one loading MS-DOS from cassette tape might have been an interesting alt-universe thing.

Among the 8-bits, the Commodore 64 was probably the cheapest around $500-600 (and another $400 for the disk drive, or $200 for the tape drive, I think).

Computers weren't something in many homes in 1984 - if you had one, it was likely an 8-bit one. PCs didn't really land in the home until later on when clones started coming down at half the price of IBM, But that was still several thousand dollars. An IBM with hard drive, single floppy and 640K was north of $5000, so a clone would likely be between $3000+ for a similar configuration.

The thing is, the MacBook Neo was done so Apple could experiment - to comply with upcoming EU laws, use up some chips they had sitting around and other things. They didn't expect the sales they got because honestly, if you survey the sub-$600 laptop market on PCs, it's rather dismal. Apple somehow packaged together something with a stunning screen, metal case and decent battery life. The CPU and RAM were middling, but for light tasks decent enough. But the display is bright and vibrant and outclasses anything you can find, the case is sold and not creaky plastic, and the battery life isn't abhorrent. It's also not a thick beast and retains the Apple aesthetic. And no stickers.

Honestly, it's something of a competitor to the iPad itself. Apple sticking a touchscreen on it and you pretty much can't justify an iPad anymore.

And the really annoying thing about this is that when the AI bubble inevitably crashes, it's going to be difficult to repurpose all of these specialized AI processors into something useful..

No, it's basically impossible. AI doesn't need high precision math - 16-bit floating point, 8 bit floating point, even 4 bit floating point is all that's required.

There aren't many uses for such low precision ALUs and such other than AI.

The border collie helps round up the cattle but it would be easier if the rancher knows where the cattle are located to send out the collie. Some ranches can be hundreds of thousands of acres. And that is the main herd. If there are stragglers, they have to be located too.

Border collies generally work with sheep, cows not as much simply because cows don't really give a damn.

But sheep you have to be careful with - an entire flock in the UK had to be euthanized because they figured how to escape their fencing, and if that knowledge spread, it would basically render all farms using that fencing to keep sheep worthless. They apparently are very good at teaching fellow sheep things like that, and it was only a matter of time before the one or two that figured out how to escape taught the rest of the flock, and that flock would then teach neighboring flocks on neighboring farms and so on. It was cheaper to euthanize the flock than to have the entire country's farms re-fenced.

The entire mission is obviously an AI fake. As acclaimed physicist Joe Rogan explained, they crew would be dead once they'd hit the Van Allen Radiation Belt (tm)

That should've been obvious by the fake globe Earth image they reportedly sent back after launching.

The earth is flat. Not a globe. The fact they had to publish this image shows the whole thing is fake.

Actually, many consumer gigabit Ethernet switches lack 10Mbps support these days. They are 100/1000baseT only.

Business and enterprise switches though I've found (including Cisco ones, which you can find dirt cheap used) still are 10/100/1000Mbps. Even newer business and enterprise class switches retain support.

Of course, once you step into 10Gbps Ethernet, you have to be careful because many only are 10Gbps only, while some do support 1/10Gbps. 2.5Gbps support is iffy unless it's specified which is annoying since many things have 2.5Gbps ports.

As for 486, there are still new CPUs using it. The Vortex86 has a CPU that executes 486 instructions though newer ones do support Pentium minus the FPU. These are modern chips, with IDE emulation of SD cards, Ethernet and USB support, as well as running at speeds of 800-1GHz.

And you've seen them used - any fast food restaurant with the ticket screens is powered by a mini PC using these SoCs. They do run Windows and MS-DOS, and early restaurant e-ticket systems used MS-DOS. But later ones nowadays use some form of Linux.

Developers don't have a culture of being economical with resources.

That's because in say, the 60s and 70s, computer time was expensive. It behooved you to make your code as efficient as possible - like today's cloud services, they often billed by the CPU cycle. And the run-debug cycle was on the order of a day, so you didn't want to make a stupid error because it meant your job got delayed a day at the least.

Sometime around the 80s and 90s, this flipped - human time was expensive. Computers were cheap and getting cheaper, RAM was plummeting as was hard drive space. The math started to work the other way - you don't want developers wasting time debugging code so libraries were popular - because it was more efficient (cheaper) to utilize the fact one person presents a well-debugged library that other developers could use and that means developers don't have to write that code, and they don't have to debug that code either.

That's why we have bloat - because it's cheaper that way. You could have a developer write nice and tight code, but how much are you willing to pay for it? If it takes them an extra week to make their library run 10% faster, was it work say, the $5-10,000 it cost? ($5000 a week is around $250K/year including benefits, or around $150K take home pay plus benefits, while $10,000 is $500,000K/year including benefits, or around $250,000-300,000 without benefits). Will that improvement let the company make back that investment?

You have to realize that if you want to charge $150K/year salary, spending a week optimizing costs the company $5000, so unless they can save that $5000 elsewhere (e.g., in reduced cloud compute fees, or customers will pay extra), there is no incentive to do it.

And that's really a valuable consideration. Also, compilers are really good these days. Like, really good. They will often so very strange things to save a few cycles. Some, like Clang, can be "too smart" and apply closed-form mathematical transforms to your computation (E.g., if you attempt to sum integers from 1 to n, and you do the "stupid way" with a loop, Clang will recognize it and actually generate the code to calculate n(n+1)/2 and eliminate the loop).

So it's a mix between the cost of a developer to optimize their code, the increasing intelligence of compilers to optimize code, and other things.

If you want to learn more about how compilers generate code, including being able to add in 0 cycles (hint: it uses the CPU's address calculation unit instead of the ALU to do simple addition and subtraction and even multiplication when it can, so the actual execution time is zero since the computation was done as part of operand calculation), Matt Godbolt of Compiler Explorer fame runs through a whole series in his Advant of Compiler Optimization series. (Youtube: https://www.youtube.com/playli... ). Trust me, it doesn't pay to outsmart the compiler.

Not a lot against the guy, but he should be #5 .. he merely continued the trajectory set by the others.

For nearly 15 years.

Jobs passed in 2011. Tim Cook has been at the helm for 15 years since then. Even if he was coasting, Apple has done remarkably well in those 15 years just coasting alone. Most companies falter and die out by then. Heck, after Apple ousted Jobs as CEO, they were struggling by the time Jobs came back and he wasn't gone nearly as long as he is now.

Even if Tim Cook did absolutely nothing for the 15 years he was CEO, the fact that Apple is still around and still going strong is already a huge credit to his (non-)leadership in managing to keep the ship steady.

Tell me how many other CEOs are like that - because history is littered with failed companies whose leadership wanted to make their mark and then their companies imploded. Like Apple nearly did 20 years ago.

Tim Cook, by "doing nothing", managed to keep Apple on the up and up, and history reveals this isn't usually what happens.

The last time, it was likely Intel buying up AMD CPUs and dumping them. As well as forcing Microsoft and Sony to use AMD chips.

AMD has its ups and downs, and Intel was riding high, but Intel needed AMD to survive to avoid antitrust activity as well. Getting rid of fabs was one thing as they are expensive.

But Chipzilla is still very big and making a lot of money. And they released new chips that are surprisingly competitive. They're not fast against the top end 9860X3D CPUs, but are very cost competitive when compared to the midrange parts, or even faster than them. At the same price or lower.

They've also seen a bump in sales in chips supporting DDR4. So going into 2025 it was doom and gloom for Intel, but for 2026 so far, Intel is holding its own and maybe even starting a comeback.

And even after all these years there are still plenty of idiots who don't understand what the word "average" means. I always see them slowing down for a camera then speeding right back up again afterwards. How do people this dumb get a license?

There was a comedy troupe that posted 3 people along the first. The first one carried a sign that read "Speed camera ahead".
The second one carried a sign that read "Just kidding".
The third one carried a sign that read "Sorry".

The thing was, there was a speed trap between the second and third person. They'd slow after the first, speed way up for the second and get caught, then the third sign apologized.

There's a YouTube video of them driving and what you'd see and it's been reported a few times.

Also, many toll roads especially in Asia do this as well - you get a ticket when you enter the toll road, then you'd hand it in when you exit and they'll charge you the toll based on distance travelled. Problem is, they also note your average speed.

Though I do remember at least one having a rest stop with a restaurant on it - so stopping for a break and/or food would lower your average speed considerably.

Honestly, that's a pretty good spread for a small crew. And is much higher than any other population sample you could take. It only happened because these things are planned out so far in advance, that Trump's anti-DEI couldn't have taken effect without delaying it for at least a couple more years.

And the moon is not an easy feat - if you want a rough idea of scale, put a half inch sized circle on one of your wrists, Then put a dot on the other wrist. Then stretch your arms wide and you have a near scale representation of the Earth, the Moon, and the distance involved. Yes, it's that far apart and likely a lot further than books and school would have you believe. Or likely you would even believe.

How are you going to host a game server on a home computer if you share your IPv4 address with other subscribers to the same ISP in the same neighborhood,[1] and the combined modem and router that your home ISP requires all subscribers to use lacks an option for port forwarding? Both of these are true, for example, of T-Mobile US Home Internet.

[1] Many home ISPs apply carrier-grade network address translation (CGNAT) to conserve IPv4 addresses since the worldwide exhaustion.

You don't. Because home internet connections, even the multi-gigabit ones, are terrible for things that need constant ping.

It's why people have been hosting services where they can get a dedicated IP, or likely one shared with similar services. Those hosting servers tend to have guarnateed connectivity.

Because direct IP connections sucked, and it's why booter services aren't so common nowadays. Because that's what happened in the past - if you got angry, you started pingflooding the host IP and making everyone's game terrible.

These days, you still have booter services, but they're services you pay for, and they're not used as much because the hosting server often hides the IP of everyone else.

I don't disagree with local servers, but maybe let's leave them on the local LAN plan. If you want to expose it to the Internet, then you can, and if you do, better have the expertise to know how to host it yourself.

And no, IPv6 won't fix this - because IPv6 only guarantees everyone gets a globally unique IP address. It doesn't guarantee that end-to-end connectivity will work. Thanks to firewalls and such which are a practical necessity these days.

The popularity of online multiplayer competitive play is driven in part by centrally hosted servers where there are people who do protect them against attacks meant to spoil the fun of everyone involved. Only a huge data center really has that ability. But if you want to host a small scale server on your own for you and a few friends, you should have the option as well

But what I think is really going on is dodgy attorneys have been putting fake citations in their briefings for centuries and we are hearing about it because they're using AI to generate the fake citations instead of just making them up on the spot like they used to. I suspect judges and defense attorneys are scrutinizing citations more too because they've seen the stories about AI.

But I do think it would be naive to believe that given how skeezy attorneys can be that they haven't been feeling their portfilings with bullshit long before AI slop was a thing

Fake citations are trivial to find. If sketchy lawyers have been putting fake citations in their briefs, you can bet it'll be found out. It's why AI generated briefs have been found out. All it would take is a lawyer competent in their area of the law, and if you cite some strange case they never heard of saying something they never heard of, they're going to look it up to see what's happening. So you might bamboozle some lawyer who doesn't know their work, but there are experts who know the case law inside and out who you can't surprise because they know all the case law and a strange citation they don't know about will be looked up just in case.

LLMs don't need fake citations to generate fake citations. All they are is a statistical text model - this token follows this token, repeat ad nauseum. Basically the AI LLM just takes some text, looks at the context to adjust the weightings of the model, then looks at say, the top 10 words that follow and depending on the settings, picks one. Then it repeats the entire process again.

Citations being fake are thus easy - the citation looks a certain way, usually a year, followed by the state or something representing the court, then a case number, and so on. And naturally those things can have weights that are random because cases span all sorts of things. Cases that have been cited more often might have a heavier weight, while lesser cited ones are less probable.