AI still needs a lot of parallelism but has low latency requirements. That makes it ideal for a large expansion card instead of putting it directly on the CPU die.

Eating the crispy chicken while thinking of the people who burned alive

I can’t imagine how you think it’s incredibly simple. These things are hell to explain to pretty much any normal person who needs to know why there’s no picture on the monitor or why their laptop/phone is not charging, or why the keyboard isn’t working in BIOS (no USB 3 support so you gotta switch to a USB 2 port). Add to that the combinatorial complexity of different cables and hubs supporting different things, and no tools for troubleshooting what feature is missing (and where in the chain) or what is suboptimal.

Worse, sometimes it’s my boss who thinks they can cheap out and get a USBC dock instead of a proper dock, forcing me to run at non-native lower resolutions or unable to use a second screen.

Search patterns yes, but also location data, and it’s aggregated over all your friends. So if you go to a restaurant together with a friend who recently searched for some clothes brand, the algorithm will know that and show you ads for that brand. Chances are you talked about his interests when you met, so you incorrectly infer that it was listening to the conversation.

Critically, “Meta, Amazon, and Microsoft told 404 Media they have no involvement with CMG’s Voice Data tool.”

But more importantly, they can’t listen on your microphone unless you give them permission. It’s not a thing that is technically possible. And like the article says, these days phones even show an indicator to alert you when the microphone is on.

Yeah I’m not sure that war crimes work that way. You don’t get a pass because the opponent is doing illegal things.

The UI looks mostly like Firefox IMO. Critically, it’s just as slow as Firefox.

This has to be a joke

Unfortunately the license agreement for .Net forbids publishing any kind of benchmark results.

WASM? Are you talking about WebAssembly?

No argument there. But apparently there’s a market for it.

More like, the devs already knew but some middle manager promised they would remove it without understanding the ramifications, and now they’ve been schooled.

The next 10 years

Is it really too much to ask that apps/devices are made secure from the ground up?

In a way, yes. They can and should definitely be made with security in mind from the ground up. But they will never be totally secure, and a necessary part of what constitutes a “secure product” is to continuously and quickly patch security issues as they become known.

Surely that’s just a secure end-to-end encrypted connection?

I would bet it’s still a bit more than that. But even if it’s just a secure end-to-end encrypted connection, here is the list of vulnerabilities fixed in OpenSSL (which is probably what they use for secure encrypted connections). It’s five so far in 2024. Then there’s some OS kernel below that which can have security issues as well. The Thermomix probably also has user authorization components and payment methods, plus various personal information that has to be protected under GDPR.

I agree that the current system is broken. So let’s say that instead of paying $300 for a pair of headphones that last three years, you pay $8.33 / month for renting the headphones. Now, if the headphones break after three years the manufacturer has to produce new ones for you. That’s an undesirable cost for them.

It is now in their best interest to make headphones that will last a long time and that they can repair if something breaks. But also, since you can easily cancel the subscription at any time, it is in their interest to offer you something that is competitive. They might even upgrade to better technology over time or add new features to the bundled app to keep you as a customer. Or alternatively, lower the subscription cost over time to reflect the relative value of the headphones.

For you, there’s also the benefit that there’s no high upfront cost that you can’t reverse. You’re paying for what you can afford in your current situation. If you lose your job you can stop paying for the headphones at a moment’s notice. I imagine that this would leave fewer people in credit card debt.

Depends what kind of ownership you’re thinking about. When it comes to electronics, “ownership” is just subscription with a longer period between payments. Your existing phone, tablet, TV, dishwasher or what have you will last a finite time and then you have to buy a new one.

If there’s something that will last a lifetime, that’s a different discussion. But those are rare. Almost every purchase you make is a commitment to a recurring cost.

Something like myfitnesspal or a thermomix shouldn’t be a subscription, there is no major updates to how someone tracks their exercise uses a hot blender that justifies it beyond users being locked in.

I won’t dispute that both of these likely abuse the subscription model for their benefit. But they definitely have a social responsibility (and in many cases a legal responsibility) to keep updating the software in these products and the network infrastructure that go with them. The internet of things is one of the most vulnerable attack vectors we have. It has been exploited many times not just to attack individuals, but to create massive bot nets that can target corporations or even countries. The onus is on the manufacturer to continuously keep that at bay. You know what they say - the “S” in “IOT” stands for security.

I know I’m in the minority but I am also a software developer, and I think subscriptions are a much healthier payment model for everyone. The issue IMO is not recurring payments but the total cost of ownership.

“Digitial goods” is very rarely just a thing that you produce once and then it’s done. The OS is regularly updated which causes incompatibilities, app stores introduce new demands, and there’s a constant stream of security vulnerabilities in your dependencies that need to be patched. Failing to adress any of these things breaks the social contract and causes rage among your users (“I PAID FOR THIS, WHY ISN’T IT WORKING/WHY AREN’T YOU FIXING BUGS/etc”). Even movies and music need to be maintained because new media formats are introduced, streaming services have to be kept responsive and up to date etc.

A subscription models the cost distribution over time much better, and it does benefit the users because it means the company can keep updating their shit even if new sales drop, instead of going bankrupt.

I don’t think this stops with just digital goods. Manufactured products (and the environment) would also benefit from a subscription model because it means there’s no incentive for planned obsolescence. It’s an incentive for keeping the stuff we already built working for a long time, instead of constantly producing new crap and throwing the old in a landfill.

But, the caveat is that this shift must not result in higher total cost of ownership for the end users over time. In fact, it should reduce the cost because repairing and updating is cheaper than building new stuff. The way many companies are pricing subscriptions today, they are being too greedy.

AI as a general concept probably will at some point. But LLMs have all but reached the end of the line and they’re not nearly smart enough.

Sure, but that’s far from a universal experience.