Oh yeah, the equation completely changes for the cloud. I’m only familiar with local usage where you can’t easily scale out of your resource constraints (and into budgetary ones). It’s certainly easier to pivot to a different vendor/ecosystem locally.

By the way, AMD does have one additional edge locally: They tend to put more RAM into consumer GPUs at a comparable price point – for example, the 7900 XTX competes with the 4080 on price but has as much memory as a 4090. In systems with one or few GPUs (like a hobbyist mixed-use machine) those few extra gigabytes can make a real difference. Of course this leads to a trade-off between Nvidia’s superior speed and AMD’s superior capacity.

These days ROCm support is more common than a few years ago so you’re no longer entirely dependent on CUDA for machine learning. (Although I wish fewer tools required non-CUDA users to manually install Torch in their venv because the auto-installer assumes CUDA. At least take a parameter or something if you don’t want to implement autodetection.)

Nvidia’s Linux drivers generally are a bit behind AMD’s; e.g. driver versions before 555 tended not to play well with Wayland.

Also, Nvidia’s drivers tend not to give any meaningful information in case of a problem. There’s typically just an error code for “the driver has crashed”, no matter what reason it crashed for.

Personal anecdote for the last one: I had a wonky 4080 and tracing the problem to the card took months because the log (both on Linux and Windows) didn’t contain error information beyond “something bad happened” and the behavior had dozens of possible causes, ranging from “the 4080 is unstable if you use XMP on some mainboards” over “some BIOS setting might need to be changed” and “sometimes the card doesn’t like a specific CPU/PSU/RAM/mainboard” to “it’s a manufacturing defect”.

Sure, manufacturing defects can happen to anyone; I can’t fault Nvidia for that. But the combination of useless logs and 4000-series cards having so many things they can possibly (but rarely) get hung up on made error diagnosis incredibly painful. I finally just bought a 7900 XTX instead. It’s slower but I like the driver better.

Speak for yourself. I’m going to migrate all of my 22-bit RSA keys to a longer key length. And not 24 bits, either, given that they’re probably working on a bigger quantum computer already. I gotta go so long that no computer can ever crack it.

64-bit RSA will surely be secure for the foreseeable future, cost be damned.

Anduril is way overengineered. I like this UI that some of my lights have:

While off:

• One push: Turn on at the last used brightness.
• Two pushes: Turn on at maximum brightness.
• Three pushes: That strobe mode that you don’t need but seems to be obligatory.
• Hold: Turn on at the lowest brightness (or moonlight mode if the light has one).

While on:

• One push to turn off.
• Two pushes to toggle between maximum brightness and the last used “regular” brightness.
• Three: That strobe mode that someone has to have some use for.
• Hold: Alternately increase or decrease the brightness.

That’s pretty easy to learn and gives you all the functions you’d reasonably need (plus that strobe) without a lot of clutter.

We used to have one: “Solang das deutsche Reich besteht wird jede Schraube rechts gedreht.” (“As long as the German Empire persists every screw is turned right.”)

Given that the German Empire failed spectacularly, this sentence isn’t very popular anymore.

Pretty much business as usual. Those things pop up every once in a while, which is why Germany has no shortage of experienced bomb disposal experts.

And yes, the fact that this is perfectly mundane is chilling if you think about it. That’s how intensely bombed areas work for decades. Battlefields don’t turn back to normal when the war is over. And we casually do this all over the world for a large number of usually stupid reasons.

They did PR campaigns against Linux and OpenOffice for quite some time – until cloud computing took off and it turned out they could earn more money by supporting Linux than by fighting it.

In fact, Microsoft weren’t happy about FOSS in general. I can still remember when they tried to make “shared source” a thing: They made their own ersatz OSI with its own set of licenses, some of which didn’t grant proper reuse rights – like only allowing you to use the source code to write Windows applications.

True, although that has happened with F/OSS as well (like with xz or the couple times people put Bitcoin miners into npm packages). In either case it’s a lot less likely than the software simply ceasing to be supported, becoming gradually incompatible with newer systems, and rotting away.

Except, of course, that I can pick up the decade-old corpse of an open source project and try to make it work on modern systems, despite how painful it is to try to get a JavaFX application written for Java 7 and an ancient version of Gradle to even compile with a recent JDK. (And then finally give up and just run the last Windows release with its bundled JRE in Wine. But in theory I could’ve made it work!)

Note that this specifically talks about proprietary platforms. Locally-run proprietary freeware has entirely different potential issues, mostly centered around the developer stopping to maintain it. Locally-run F/OSS has similar issues, actually, but lessened by the fact that someone might later pick up the project and continue it.

Admittedly, platforms are very common these days because the web is an easily accessible cross-platform GUI toolkit SaaS is more easily monetized.

And this is why stuff should be defined in terms of day’s earnings to provide scaling. If an ultra-rich person gets jailed and has to post 20 billion dollars in bail, they can’t treat jail as a minor inconvenience.

Go full Lovecraft and spell it “yog-huurt”.

Nope, they just become less predictable. Which is why in some parts of Germany you can’t build as much as a garden shed without having EOD check the land first. In the more heavily-bombed areas it’s not unusual to hear on the radio that you’re to avoid downtown today between 10 and 12 because they’re disarming a 500-pound bomb they found during roadwork.

And yes, the fact that an unstable bomb capable of trashing a city block is mundane nicely illustrates war’s potential to fuck things up for generations.

Japan might want to get that land under and around the airport checked. There might be some other surprises hidden down there.

The main issue was that Vista asked for admin rights all the time. One of the first things they addressed with SP1 was to require admin privileges for fewer operations, cutting down on the number of UAC prompts.

My reaction exactly. I studied there as well. Lise Meitner may be underappreciated but at least someone made sure she’s not forgotten.

True. Just this weekend I spent far too much time trying to get a printer to work again on Windows after its IP address got changed. In the end Windows refused to talk to the printer unless I removed and then readded the device from the Settings app, which prompted a reinstallation of the device driver. No, just changing the IP address in the device settings wasn’t enough; Windows insisted on the driver being reinstalled.

Linux didn’t need reconfiguration; it just autodetected that the printer had moved.

I’m not saying that Linux is without issues, not by far. But Windows has never been terribly “it just works” for me either. The closest to “it just works” was (aptly) OS X somewhere around Snow Leopard.

You forgot the degaussing sound for those screens that had that feature. Like turning them on but louder.

*KLONK*

Bonus points for not turning your parents’ backyard into a Superfund site.

And I wouldn’t know where to start using it. My problems are often of the “integrate two badly documented company-internal APIs” variety. LLMs can’t do shit about that; they weren’t trained for it.

They’re nice for basic rote work but that’s often not what you deal with in a mature codebase.

Like every time there’s an AI bubble. And like every time changes are that in a few years public interest will wane and current generative AI will fade into the background as a technology that everyone uses but nobody cares about, just like machine translation, speech recognition, fuzzy logic, expert systems…

Even when these technologies get better with time (and machine translation certainly got a lot better since the sixties) they fail to recapture their previous levels of excitement and funding.

We currently overcome what popped the last AI bubbles by throwing an absurd amount of resources at the problem. But at some point we’ll have to admit that doubling the USA’s energy consumption for a year to train the next generation of LLMs in hopes of actually turning a profit this time isn’t sustainable.

Depends. On Linux or older macOS where light mode typically means a comfortable light gray? Light mode is the way to go. On Windows where light mode means an eye-searing onslaught of #FFFFFF? Dark mode is the only sensible choice.