Ahh, okay. I think the main confusion is I was kind of ignoring the code portion of the library, which is easily deduplicated like you said.

Any allocations done by the library running (opencl.so in this example) will still be separate per process though, unless the program is passing mutable shared memory buffers around, which could be the case if there’s IPC going on.

The point about memory usage being hard to measure is definitely true, especially since linux won’t actually consume memory until a write is made to the block, unlike Windows which preemptively allocates the full requested size.

While shared memory is a feature in linux, it’s mostly used for inter-process communication and isn’t something that happens by default with shared libraries.

When a program loads a shared library, it creates a new instance of it that runs in the same memory space as the program.

In the example you’re using with KDE, that’s actually a system process that your program is communicating with, so it’s not quite the same as just loading a shared library (.so). Since new programs can connect to the existing system service, that can end up saving RAM by having a single system program service multiple user applications.

This isn’t really any different on Windows where the desktop environment (dwm.exe) is handling drawings and compositing for all user apps.

Edit: Based on the downvotes, maybe I got something wrong here. Maybe someone can comment on what? I’m happy to learn

It’s only clean when you compare it to the fossil fuel alternatives it replaces like coal.

How about…

• ICE kidnapping citizens off the street
• DOGE dismantling government services
• The blatant scams and rug pulls Trump is pulling to rob taxpayers
• Companies getting billions of tax breaks to build AI datacenters nobody wants

I could go on…

Seriously? Flock is the only problem you have with America right now? You must be living under a rock.

How nice of the world’s computers to all communicate the time based on how old Jolteon and Jolteon Jr are.

I think you’re confusing your units for power vs energy. 3.5GWh (Gigawatt-hours) says nothing about the size of a power plant without also providing a time period that energy was generated in. You likely mean 3.5GW (Gigawatts), which is a measure of instantaneous power (in this case peak power). A 3.5GW plant generates 3.5GWh in 1 hour.

I think you’re missing the space after > for the quote syntax.

Like this

With how big the cab is, it almost looks like they took an SUV and bolted on a tiny truck bed sticking out the back. The proportions are very strange.

It’s got 50% more range, so it’s probably got a bigger battery. Lithium is expensive.

Unless you’re renting a GPU instance, I highly doubt your VPS is in one of these 1GW AI facilities, they’re mostly used privately by companies like Anthropic and OpenAI, or other companies specifically renting AI training capacity.

AWS has had existing datacenters for decades now and while they’re quite large, they’re a fraction of the size the AI industry is demanding. They were also built before regulations started getting changed in favor of more datacenters, so they’re actually following the zoning rules about noise, water, and power consumption.

The type of datacenters running lemmy and other services are nothing like the AI datacenters being built. Some of the AI datacenters are literally 1000x bigger in terms of power consumption than existing data center buildings are using. They’re trying to build a 10 Gigawatt AI datacenter in Ohio, and it will need its own power station to run it.

I own my server and colocate it, and it’s sitting in what looks like a normal office building. You would never know from the sidewalk directly outside that there’s a hub for AWS and Google Cloud right there. The amount of cooling required is small enough they can just use regular air conditioning units on the roof, and none of this evaporating water crap.

This being partially right is probably exactly how it ended up in the training data. But AI is too dumb to understand which parts are relevant/important.

Polarization is kind of just magic. Which means these sunglasses might actually require mana points to use.

Well, it really depends on how the notification is sent. Most people don’t have push notifications set up on YouTube, so the spike wouldn’t be crazy high there. Even a few 1000 people requesting a video at once could overwhelm a single server’s bandwidth though.

I completely guessed at the 2% number if they were on another platform that sent out emails or push notifications. But I’m sure they’d also lose a huge chunk of those subscribers when moving platforms.

I can get behind planting more trees, but it sounds like they just started planting them on city-owned land without permission? I somehow doubt an argument with a cop about how great the trees will be is going to get them out of that.

25M subscribers means 25M people getting notified to come watch a video all at once. They could easily end up with a half million requests to buffer the new video all at once, and that would take down anything short of a full CDN to handle local geo-caching. It wouldn’t be cheap to handle the burst loads of new releases.

It just means they’ve been caught again.

"intuitively find and replace non-identical things based on what i just did”

We had that 5 years ago with GitHub Copilot, and things have only gone downhill from there…

I am pretty much forced to use AI at work, and I still come home and write code by hand because I enjoy it. A good language server is much more helpful than the LLM autocomplete if you already know what you’re trying to write.

Your image works fine for me. Idk what they’re talking about, these companies are making plenty of profit. (Probably not enough if you ask the shareholders tho)