To be fair, having a single snake-like cable connecting all the PCs was a pretty common idea. To us today it seems obviously easier to have individual connections to a central hub or switch, but at the time a common cable was accepted, perhaps to save on the total amount of wiring used. Token Ring also used a single wire, though joined into a loop, and I think there was also Token Bus.

You are right that "transmit, and retry on collision" can never scale. But it is simple and hard to get wrong, as long as everyone is polite enough to wait a little before retrying. I'd say the success of the original Ethernet is an example of "worse is better".

I thought that looking moderately downward was the best position for your neck. Hence the advice to keep the top of your screen at eye level. (I don't quite follow that: my screens are all in portrait orientation and the top of each display is slightly above my eyes, but for most of the screen area I'm looking level or downwards.) I guess there's a sweet spot where you are looking downwards, but not too far down?

Digital security experts told me that bad guys can use software to easily translate your âoeatâ and âoedotâ into a regular old email address.

Well, duh. It's trivial. But the question is not whether they can, it's whether they do. The evidence, as far as I can tell, is that generally they do not. It's hardly worth it for them as anyone with the moderate level of intelligence needed to obfuscate their email address is too intelligent to respond to spam. The exception would be if some large, popular website started displaying users' email address using an automatically applied obfuscation. But I think sites aren't that stupid - they usually have "click to reveal the address", which requires a new request to the server and rate-limits any harvesting.

Ah, but apparently this might be a limitation of Parallels Desktop, and not a fundamental limitation of the x86_64 architecture, since

Windows 365 supports nested virtualization...

How does Microsoft manage to do it while Parallels cannot?

Microsoft does say that there are some limitations when it comes to virtualizing Windows 11 on top of macOS, pointing primarily to features that require nested virtualization to function as not being supported.

And that, I think, is the problem with operating systems like Qubes OS, or indeed Windows 11 itself with its Android and Linux subsystems. The PC cannot be fully virtualized, because the virtual environment cannot itself host VMs.

I have no experience on mainframes but I believe that where virtualization is done properly, it's not like that -- a virtual machine can itself host other VMs, down to an arbitrary depth. Anyone care to correct me?

Penetration testing is fine, but like most testing it can only prove the presence of bugs, not their absence. If you are defending against highly skilled attackers (and let's assume for the moment that Russia or North Korea or China or whoever has such people, and might want to target your oil pipeline or reactor or whatever) then your penetration testers need to be equally skilled. That doesn't come cheap. You can't rely on some box-ticking exercise where a moderately intelligent monkey connects to your network and sends a few packets of random data to see what crashes.

The Sassy Justice video has some very good voice impersonations that aren't cartoony. You have to skip a few minutes of filler to get to them.

I was most interested in Linus's last remark:

That way we could finally get rid of CONFIG_MATH_EMULATION too.

I guess that's true, Intel never made a cut-price Pentium chip with no floating point unit, and I guess none of the competing vendors did. But isn't it possible we might get CPUs without x87 floating point produced in the near future? Since if you have high-performance floating point code you will be using SSE or later instructions, and crusty old stuff calling x87 will still run acceptably fast under emulation.

Perhaps you will say that silicon is so cheap it costs basically nothing to include a small x87 floating point unit on the die?

No, these are well established funds that make up the "quant trend following" sector. Most have been around for a couple of decades or more. It's not a case of putting them into a category post hoc. Of course, in any year there will always be some sector or some strategy that has surged ahead...

I think Apple gives slightly more courteous treatment to a big firm like Microsoft than they would to a small-time app store developer.

Well, windows can have mosquito netting. Doors need to let through people but not mosquitoes. I would like a ventilator above the doorway that blows air downwards whenever the door is open. Mosquitoes cannot fly in strong wind and, I think, they will keep away from it.

If you can compress a document (be it text, image, or sound) to a summary of its key attributes, then recreate it, then in a sense you have understood it. In a very loose sense - it's not a viewpoint I fully agree with - but until philosophers come up with a generally agreed, clearly defined definition of "to understand" then a definition based on compression and Kolmogorov complexity (the smallest computer program that generates the document) is as good as any other.

This viewpoint motivated the Hutter Prize, offering a prize to the first to compress the one gigabyte corpus to less than 115 megabytes (including the size of the decompression code). "This compression contest is motivated by the fact that being able to compress well is closely related to acting intelligently, thus reducing the slippery concept of intelligence to hard file size numbers."

Here it's a bit different since you have a lossy compression scheme - the exact input image is not returned. That means you couldn't really offer a contest with a cash prize, since the judges couldn't fairly decide whether the output was "close enough". Nonetheless you could say that the model has demonstrated some kind of intelligence in being able to recreate a close-enough image from a short description. As usual with machine learning, it's hard to get the model to explain how it produces the result or what key features it has discerned.

That's not quite fair. If you live in the USA and you wanted to pay someone in US dollars, but they lived in a different state, you'd have to drive there and hand over the cash. It would be even slower and more expensive if they lived in another country.

In practice it's fast because you are not paying in physical currency, but in a kind of electronic promise made by banks. Your bank balance falls, the other person's increases, and everyone trusts that in the end you will be able to convert the electronic balance back to physical dollar bills if you want.

The cryptocurrency world has largely failed to develop this kind of intermediary layer. That's partly because of terrible security -- would you really trust one of these bitcoin deposit sites not to get hacked? and partly because to have banks, you need some authority making sure they have enough physical reserves to back up the balances they create. That could be a full reserve requirement, so the bank holds bitcoin equal to at least 100% of its liabilities, or it could be fractional reserve banking, where only a part must be held. If the latter, you need a central bank and some kind of deposit guarantee scheme to prevent bank runs, but there is no central bank of bitcoin.

The ROM was actually a little bit slower than RAM, so you could speed up your machine by a few per cent by copying the ROM to RAM and remapping the address space. I think most home computers of the 1980s were instant-on. There's nothing remarkable about switching on a C64 and it showing the prompt immediately. What's unusual about the Archimedes is that it maintained the "home computer" feel (both good and bad) into the 32-bit era.

The original RISC OS 2 booted to the desktop in about a second. Later versions became slower. Then if your machine had a hard disk there was the temptation to make it load all sorts of stuff on startup. Mine became so bloated I made it play music and colour cycle the background while things loaded.