I don’t think calling hallucinations a bug is strictly wrong, but it’s also not working as intended. The intent is defined by the developers or the company, and they don’t want hallucinations because that reduces the usefulness of the models.

I also don’t think we know that it is a fact that this is a problem that can’t be solved in current technology, we simply have not found any useful solution.

It is generally considered a bad idea to use envs for passing secrets in general since envs for process n are available to other processes which have access and permission.

Comparing python to rust, rust has far fewer breaking updates than python, and thats a fact. Feature updates can and do break older code in python, whereas in rust this is simply not allowed with few exceptions.

The language is allowed to change in compatible ways with editions. Every few years a new edition is released which allows otherwise breaking changes to be implemented, but the old and new code can still work together. Developers can rev the edition version when they want. I also think cargo might be able to help upgrade to a new edition as well.

Rust isn’t perfect, but python fails to learn the lessons that even perl implemented decades ago.

To be honest, I never heard of it, and it is interesting, but the language isn’t the only factor, it’s the ecosystem as well. It says it’s an alternative to C, so I will just assume it can consume C libraries. But that still leaves you with using C libraries, which is not a great position to be in if you are looking to not use C.

If you are looking for something that is actually in use, but not rust, look into Zig. Still would need to use a lot of C libraries, but it at least looks like it has momentum. Not to mention they seek to completely replace libc, which would actually be useful and an achievement, since that is the biggest problem C actually has.

I am a rust fan myself, but if you are new to programming it’s not a great place to start due to its’ learning cliff.

Maybe, but i never mentioned years into the future. Of course technology will improve. The hardware will get better and more effcient, and the algorithms and techniques will improve.

But as it stands now, i still think what i said is true. We obviously don’t have exact numbers, so i can only speculate.

Having lots of memory is a big part of inference, so I was going to reply to you that prices of memory stopped going down at a similar historical rate, but i found this, which is interesting

https://ourworldindata.org/grapher/historical-cost-of-computer-memory-and-storage?time=2020..latest

The cost when down by about 0.1x from 2000 to 2010. 2010-2020 it was only about 0.23x. 2020-2023 shows roughly another halving of the price, which is still a pretty good rate.

The available memory is still only one part. The speed of the memory and the compute connected to it also plays a big part in how these current systems work.

Of the things people complain about that systemd brings in, this is among the least offensive. It makes sense for an init system to provide such functionality, the function of spawning new system processes.

Additionally, in modern systems it doesn’t make sense to use such features. Spawning a new process per request or on demand doesn’t gain you much and does reduce performance.

Spawning new processes on most OS is pretty slow compared to other operations. Additionally, there would also be an increase in latency as the new process needs to be loaded, whereas most software these days can handle the new request in more efficient ways.

I think you can also try to reuse the same process for multiple requests, stopping it only once it has been quiet for a while. But this still doesn’t really help much.

Historically, i think it was used to try to save memory. But today its a bigger nusance than it is worth. I just checked how much memory sshd is using, and i think it is less than 10mb.

total kB 8508 6432 1160

And to be clear, you theoretically can’t save much if any memory doing this because you must have enough memory available to be able to run the process, otherwise bad things will happen or some other process gets oomed.

Additionally, spawning a new process per request can represent an availability violation. An attacker could launch a series of very slow connections to a server spawning a new process per request, causing a depletion of resources.

With all that said, I wouldn’t say there are no uses at all for this, it can be useful to make very minimal network connected software that does some very basic stuff in a secure network.

If the product costs that much to run, and most users aren’t abusing their access, it’s possible the product isn’t profitable at any price that enough users are willing to pay.

That’s fine and easy on desktop/web browser, but for mobile devices it is not quite as easy. You would either need to use a hacked version of the app or a third party app.

Most users likely do not know about recall, and as the guy in the video shows, there doesn’t appear to be anything in a normal user interface showing that it is installed and configurable.

If you don’t want to risk getting a ban at all, the only safe thing is to not connect to the internet at all. Maybe there is some level of safety, but it could take only one mistake.

If we assume that we fully understand how nintendo catches this, we would still only ubderstand at that point in time. They could still change or push updates which could cause you a problem.

A ton of people using github barely understand the different between github and git and often think they are the same thing or that github and git are somewhat related more than they really are.

The last I looked into it, the best way to do it was to get an older kindle so you could download the older DRM copies of books from amazon. But I think some newer books are using only the newer DRM which I don’t think has been cracked.

It has probably been at least a year since I checked. If you do end up finding an updated method, I would be interested.