That requires an unlocked phone. If the phone is locked it will reject the eSIM from an operator that the locking operator doesn't want to allow.

Just because one thing is true, doesn't mean the other thing can't be partly true. I also benefit from the eSIM for similar reasons, but the article is clearly pointing out a serious problem with them and that problem is not really just by accident. Operators long disliked the physical SIM card for a bunch of reasons and the eSIM addresses a bunch of the things mobile operators hate; inventory, techinical support of the SIMs and so on.

I'd like to know what the solution is for people who have eSIMs. It seems to me that it becomes critical to have access to the operator's support portal using a password or other token which is completely separate from the phone so that you can renew your eSIM without needing the SIM working. That can be a big problem for portals that authenticate using the SIM, so lots more care is needed by the user.

I think you have misunderstood. The aim here is not to predict the future, it is to create it. Everyone should understand this; "AI" as we now know it has no original thought other than limited interpolation made interesting through the injection of randomness. All "AI" is "stealing" or "unlicensed copying" other people's works in order to make money from them without having to do the effort of actually learning* about and understanding the topic that the AI generates content about.

These guys are trying to make their "AI" systems invaluable and heavily embedded into all sorts of important services. Their aim is to

There are two points of view here. The first is the copyright point of view. The AI database / "neural network" / "model weights" is what copyright people would call a "derivative work" of the works that are in it. In particular, because important and valuable copyrighted works are repeated all over the training set by default it would be very easy to generate complete copies of them from the database if it weren't for special filters that AI people use to hide that. The overall effect of this is that, because judges have not understood that model weights are a derivative work of all of the works used to train the model to a greater or lesser extent copyright no longer works. To fix this you would need to make three changes.

* models would need to be officially and clearly recognized as derivative works
* filtration to hide works from database output would need to be recognized as proof of willful infringement
* actual output matching fragments of content would need to be metered and covered through licensing

The alternative point of view currently simply redefines the meaning of "derivative work" so that model weights are not counted as such or at least come under some form of "fair use" exception. This is the view that AI venture capitalists are trying to push. They want the governments to allow them to take for free what they would otherwise have to pay for. However, the problem is that they are then attempting to own and control the output of their own models - most importantly wanting to have their own copyright or trade secrets protecting their model weights. For this to work you would need changes, for example:

* people need to be given a fair reward for their work
* valuable work which needs to be done needs to be identified for such support
* ownership of model weight data needs to clearly belong to the group of people that provided the work which allowed that model to exist.

If we want to limit the extension of copyright but still have some reward for creators, we need some way to ensure that those VCs that are benefiting from the content which the copyright system has created for them then pay back. Doctorow is proposing that this could be treated as a workers rights issue. For example, if someone's work is used in creating an AI, they are guaranteed a minimum wage or negotiation rights for payment from the people who are benefiting from the AI.

The VCs are out trying to get their products into everything whilst trying to avoid the responsibility which should come from the content and work of others that they have used. Understand that and you should demand that they pay for what they take. One way or the other.

* AI people use the term "learning"; part of understanding that is needed here is that learning is not the same thing as you our I mean by learning which requires understanding. It is more equivalent mere rote repetition or memorization.

This is true, but Chinese cars sold in Europe are not nearly as cheap as Chinese cars sold in China. Still, they are often decisively cheap.

They sold their shares, moved to Hawaii and will never care about your opinion (or that of anyone else in the computing industry).

Xorg has been forked and active development is now done at Xlibre. Hopefully that means that there will be enough pressure on Wayland that they actually fully reimplmenent all X functionality that people need. In any case, it provides a clear backup.

If Rust is to have a good future it's time for them to learn from the Python people. Python has a bunch of wierd stuff; functional programming; comprehension, introspection and so on going on. However the main good thing that they do is being slightly embarrassed about the complexity.

Look at the language disasters - C++, Perl and perhaps even Scala, for example. What lots of them have in common is that they boast about their "sophistication" - it's summarized in the Perl motto "there's more than one way to do it". Compare that with the Python

"There should be one—and preferably only one—obvious way to do it,"

(my emphasis on obvious - this is not about lack of choice - it's about making it easy to decide)

What the fat that there's one obvious way to do it means is that if you select the obvious way then 90% of other people who do the same thing will also do it in the same way you do which means that when something goes wrong you are not alone and normally someone else has already solved your problem and even if they haven't, they are interested in solving the same problem and will work with you to help solve it together.

When you have infinite choice and no obvious correct solution, like in C++ or Perl, you end up with a unique solution that nobody else is using and if you have a problem then you don't get any of the benefits of F/OSS collaboration. That's a disaster.

The Starlink Dish isn't moving. It's an active array and it's completely possible for that to either "point" in two directions simultaneously or to time switch so that it measures first one satellite then moves to the other.

a) The fact that Starlink controls the software on the terminals and is allowed to send the location from the terminal to their system.

b) The fact that Starlink clearly states that they know, and need to know, your exact terminal location in order to allow handovers to happen correctly.

Teach your grandmother to suck eggs. Some phrases for you to look up to start to dig yourself out of the hole of your ignorance.

"timing advance"
"round trip time"
"propagation delay"
"triangulation"

You might study how Malaysia Airlines Flight 370's flight path has been investigated to get some popular and simplified explanations that will help you with imagining how this could be done.

In the meantime in this house we follow the laws of science and the experiment is in. 8m location accuracy has been demonstrated with Starlink so you are just fundamentally wrong. Once the experimental evidence contradicts your theories, that's the point at which you start trying to learn.

Just wrong. The spacex satellites themselves provide location independent of GPS and 8 m accuracy has been demonstrated even without SpaceX cooperation. SpaceX themselves could probably achieve 1m accuracy if they wanted to, but in any case 50m accuracy would be enough for target location.

Because you are a Russian disruption agent and the idea of an independent judiciary and trial by jury who would never convict someone because the president wanted them to is completely alien to you. There's plenty wrong all over American politics and nobody should deny it, but going after the innocent is one of the best ways of stopping people going after the guilty.

SpaceX knows exactly where each terminal is because they have to for radio timing to work correctly with the moving satellites. Russia has to test those terminals somewhere. SpaceX could be giving that location to Ukraine. SpaceX could be permanently destroying any terminal that turns up near there and hasn't been cleared by Ukraine. SpaceX could be reporting the incoming locations of missiles and could be cutting service as soon as they realize that there's a surprise terminal moving rapidly towards a Ukrainian city.

Due to their lack of accuracy which makes them ineffective against hardened military targets, most of the Russian missiles are used in strict terror campaigns against civilian buildings: Power stations providing neighborhood heating, residential tower blocks with hundreds of families living in them, independent churches, nurseries often at times of maximum use and of course hospitals.

Getting on top of this and ensuring that Ukraine is a key supported customer that feels it gets what it needs could have allowed a real feel good story showing a company that took abuse and murder seriously. In fact it's pretty clear that some working at Starlink tried to do that at times and the management made it more difficult. Elon Musk is literally a baby killer.

Were you able to before? Obviously, "like everyone else" you'll be keeping it encrypted at rest with keys that are kept in an HSM. For the important "pet feeding habit data" you will have made an exception and actually bought your own HSMs, kept in your multiple highly geographically separated underground bunkers with limited on site compute and simply feed limited summary results back to the cloud. For less important "nuclear weapons test results" data you find some compromise where you can track which and for the "current location of warheads data" you might just decide that you take the risk because, you're just going to have to accept the risk of that data leaking anyway for other reasons, and cost of processing is a priority. That's why right data classification and appropriate handling is important.

Every computer you use you trust Intel, Infineon, Samsung and tens or hundreds of others manufacturers not to embed hostile radio devices or software on chips that phone home. That includes some companies that actually have added weird management systems against their users wishes. Apple (IIRC) have pubilcly had the situation where they threw away a bunch of mother boards that arrived for their cloud. I'm sure all of the others have too. What don't they spot?

Adding one more cloud provider doesn't really make things much worse that it was to begin with.

Generally a few things

1) eminent domain for countryside land
2) tunnels into cities.
3) once the network is in existence cities that don't have it lose out and will make a big effort to find land for

The distances in America are much bigger, so ideally you'd move to a faster rail standard, either simply double European width or maglev. Of Underground tunnels, though, are a really big thing because the main benefit of trains is that you can run them right to the center of the city so that people living there can leave their offices 20 minutes before the train, walk or take a taxi, get to the station 10 minutes before departure and still safely get their train.

At the other end it's even faster because you don't need the 10 minutes of leeway.