A major part of that is, I think, that desktop OSes are, “by default, insecure” against local software. Like, you install a program on the system, it immediately has access to all of your data.

That wasn’t an unreasonable model in the era when computers weren’t all persistently connected to a network, but now, all it takes is someone getting one piece of malware on the computer, and it’s trivial to exfiltrate all your data. Yes, there are technologies that let you stick software in a sandbox, on desktop OSes, but it’s hard and requires technology knowledge. It’s not a general solution for everyone.

Mobile OSes are better about this in that they have a concept of limiting access that an app has to only some data, but it’s still got a lot of problems; I think that a lot of software shouldn’t have network access at all, some information shouldn’t be readily available, and there should be defense-in-depth, so that a single failure doesn’t compromise everything. I really don’t think that we’ve “solved” this yet, even on mobile OSes.

Bind mounts aren’t specific to Docker. You’re asking specifically about bind mounts as used by Docker?

I use, say, bash quite happily. But I will also come down pretty firmly on the side of static typing for large software packages. It lets software handle a bunch of rigorous checking that otherwise eats up human time.

I can believe that LLMs might wind up being a technical dead end (or not; I could also imagine them being a component of a larger system). My own guess is that language, while important to thinking, won’t be the base unit of how thought is processed the way it is on current LLMs.

Ditto for diffusion models used to generate images today.

I can also believe that there might be surges and declines in funding. We’ve seen that in the past.

But I am very confident that AI is not, over the long term, going to go away. I will confidently state that we will see systems that will use machine learning to increasingly perform human-like tasks over time.

And I’ll say with lower, though still pretty high confidence, that the computation done by future AI will very probably be done on hardware oriented towards parallel processing. It might not look like the parallel hardware today. Maybe we find that we can deal with a lot more sparseness and dedicated subsystems that individually require less storage. Yes, neural nets approximate something that happens in the human brain, and our current systems use neural nets. But the human brain runs at something like a 90 Hz clock and definitely has specialized subsystems, so it’s a substantially-different system from something like Nvidia’s parallel compute hardware today (1,590,000,000 Hz and homogenous hardware).

I think that the only real scenario where we have something that puts the kibosh on AI is if we reach a consensus that superintelligent AI is an unsolveable existential threat (and I think that we’re likely to still go as far as we can on limited forms of AI while still trying not to maintain enough of a buffer to not fall into the abyss).

EDIT: That being said, it may very well be that future AI won’t be called AI, and that we think of it differently, not as some kind of special category based around a set of specific technologies. For example, OCR (optical character recognition) software or speech recognition software today both typically make use of machine learning — those are established, general-use product categories that get used every day — but we typically don’t call them “AI” in popular use in 2025. When I call my credit card company, say, and navigate a menu system that uses a computer using speech recognition, I don’t say that I’m “using AI”. Same sort of way that we don’t call semi trucks or sports cars “horseless carriages” in 2025, though they derive from devices that were once called that. We don’t use the term “labor-saving device” any more — I think of a dishwasher or a vacuum cleaner as distinct devices and don’t really think of them as associated devices. But back when they were being invented, the idea of machines in the household that could automate human work using electricity did fall into a sort of bin like that.

I like scrambled eggs to omelette.

You probably want “I prefer scrambled eggs to omelette”.

Scrambled eggs if I’m making it. Slightly less effort. If someone else is, I guess maybe omelette. Not really a strong preference.

You mean a dedicated hardware device?

This isn’t a huge increase. It’s not on the order of what just happened with RAM.

According to a report from Digitimes Asia (quoting Nikkei), HDD contract prices jumped roughly 4% quarter over quarter in Q4 2025.

I mean, you can sue for whatever. Doesn’t mean that you’ll win or that the court won’t throw it out.

smart TVs

Instead of getting some proprietary media device glued to a display, you could just put Kodi or similar on an HTPC. Use whatever display device you want. That way, you decide what the device does.

No place like 0177.0x1.

My hypothesis is that’s the pear-shaped Army chick that lives across the street, but it might be the balding middle aged family man two doors down.

If you can wander around with kismet and a GPS sensor on a laptop, you can map the location of a WAP.

Women’s clothes tend to be more prone to vanity sizing than men’s.

Vanity sizing, or size inflation, is the phenomenon of ready-to-wear clothing of the same nominal size becoming bigger in physical size over time.

Vanity sizing is a common fashion industry practice used today that often involves labeling clothes with smaller sizes than their actual measurements size. Experts believe that this practice targets consumer’s preferences and perceptions.

If you mean distributing inference across many machines, each of which could not individually deal with a large model, using today’s models, not viable with reasonable performance. The problem is that you require a lot of bandwidth between layers; a lot of data moves. When you cluster current systems, you tend to use specialized, high-bandwidth links.

It might theoretically be possible to build models that are more-amenable to this sort of thing, that have small parts of a model run on nodes that have little data interchange between them. But until they’re built, hard to say.

I’d also be a little leery of how energy-efficient such a thing is, especially if you want to use CPUs — which are probably more-amenable to be run in a shared fashion than GPUs. Just using CPU time “in the background” also probably won’t work as well as with a system running other tasks, because the limiting factor isn’t heavy crunching on a small amount of data — where a processor can make use of idle cores without much impact to other tasks — but bandwidth to the memory, which is gonna be a bottleneck for the whole system. Also, some fairly substantial memory demands, unless you can also get model size way down.

I’d also note that OP only created their account a day ago.

I wonder how much exact duplication each process has?

https://www.kernel.org/doc/html/latest/admin-guide/mm/ksm.html

Kernel Samepage Merging

KSM is a memory-saving de-duplication feature, enabled by CONFIG_KSM=y, added to the Linux kernel in 2.6.32. See mm/ksm.c for its implementation, and http://lwn.net/Articles/306704/ and https://lwn.net/Articles/330589/

KSM was originally developed for use with KVM (where it was known as Kernel Shared Memory), to fit more virtual machines into physical memory, by sharing the data common between them. But it can be useful to any application which generates many instances of the same data.

The KSM daemon ksmd periodically scans those areas of user memory which have been registered with it, looking for pages of identical content which can be replaced by a single write-protected page (which is automatically copied if a process later wants to update its content). The amount of pages that KSM daemon scans in a single pass and the time between the passes are configured using sysfs interface

KSM only operates on those areas of address space which an application has advised to be likely candidates for merging, by using the madvise(2) system call:

int madvise(addr, length, MADV_MERGEABLE)

One imagines that one could maybe make a library interposer to induce use of that.

I mostly use terminal-based software on Linux.

I think that the only programs I use much that embed a web browser are:

• Firefox

• Steam

• Some games that are Web-based and which I only run one of at once (Neo Scavenger, some RPGMaker-based games, probably some others).

In most respects, I’d rank a five-year-old human mentally above a raven or octopus.

https://en.wikipedia.org/wiki/Cannibalism_in_Oceania

The Korowai tribe of south-eastern Papua could be one of the last surviving tribes in the world engaging in cannibalism.

Sometimes we do.

In the US, looks lile it’s illegal to trade in dolphin meat due to the Marine Mammal Protection Act.

It’s apparently legal in Japan.