Where is the website template from? I’ve seen the exact same one before.

You can run your own AI locally if you have powerful enough equipment, so that you’re not dependent on paying a monthly fee to a provider. Smaller quantized models work fine on consumer-grade GPUs with 16GB RAM.

The major issue with AI providers like Anthropic and OpenAI at the moment is that they’re all subsidizing the price. Once they start charging what it actually costs, I think some of the hype will die off.

I definitely agree with you!

I’m using AI a little bit myself, but I’m an experienced developer and fully understand the code it’s writing (and review all of it manually). I use it for tedious things, where I could do it myself but it’d take much longer. I don’t let AI write commit messages or PR descriptions for me.

At work, I reject AI slop PRs, but it’s becoming harder since AI can submit so much more code than humans can, and there’s people that are less stringent about code quality than I am. A lot of the issues affecting open-source projects are affecting proprietary code too. Amazon recently had to slow down with AI and get senior devs to review AI-written code because it was causing stability issues.

… did you read the same article as everyone else? I can’t tell if you’re joking or not.

I think the blurb was posted by the submitter (@[email protected]) rather than being a part of the link.

If your AI is making PRs without you, that’s even worse.

This is happening a lot more these days, with OpenClaw and its copycats. I’m seeing it at work too - bots submitting merge requests overnight based on items in their owners’ todo lists.

for example maybe AV1 takes even more off,

I know this was just an example, but Intel 11th gen and newer has hardware acceleration for AV1.

GPUs have their place, but they significantly increase power consumption, which is an issue in areas with high power prices.

If you want to self-host email or websites, I’d use a VPS for those use cases. For websites, a $30/year VPS would be more than sufficient. You can try host at home, but hosting those things from a residential IP doesn’t always work well.

QuickSync is more than sufficient for most users. It can handle several concurrent 4K transcode. It’s also not that common to have to transcode, unless you stream your media content when away from home a lot, and have poor upload speed.

If going Intel, there’s different models of Intel iGPU, so I’d go for the lowest-end GPU that has the higher end iGPU. My home server is a few years old and has an Intel Core i5 13500. The difference between the 13400 and 13500 looks small on paper, but the 13400 only has UHD Graphics 730 while the 13500 had UHD Graphics 770 which can handle double the number of concurrent transcodes.

Intel iGPUs also support SR-IOV which lets you share one iGPU across multiple VMs. For example, if you have a Plex server on the host Linux system, and Blue Iris in a Windows Server VM, and both need to use hardware transcoding.

I’ve heard AMD’s onboard graphics are pretty good these days, but I haven’t tried AMD CPUs on a server.

You can share the node with them, and use an ACL to control which ports they have access to.

The computers they were throwing away were broken, and they didn’t have a use for PC-133 RAM any more.

Those are some real late 2000s hinges on the laptop in the photo.

I would have loved to have 32 MB RAM. I was stuck with a 486 with 16 MB RAM and 600ish MB HDD until 2003 or so, because we couldn’t afford to upgrade. I think I upgraded to a second-hand Pentium 3 at that point, and upgraded the RAM with mismatched RAM modules (different brands, different capacities) salvaged from systems my school was throwing away.

A simpler time. I miss it sometimes. Neither me (as a teenager) nor my parents had any money, but I did have enough free time to learn how to code and play shareware games. It gave me something to do that didn’t cost much money. Over 20 years later and I’m still coding.

At least this looks like it actually tried to do something.

There was similar software for Windows, called SoftRAM. Turns out it didn’t actually do anything. Their driver was just sample code from Microsoft, and the app reported fake RAM savings.

This is so cursed that I want to try it out.

I used to use Dogpile a lot in the late 1990s. Coincidentally it was a similar idea to this and SearxNG - it was a meta search engine that combined Yahoo, Lycos, Excite, AltaVista and a few others into one interface (no Google since it wasn’t in widespread use yet).

They have a free trial account that lets you do 100 searches. It’s worth trying out at least.

How does this differ from every other distribution method, though? You can just as easily do something malicious with an Appimage or Debian/rpm package.

Have you tried Kagi?

If you want to share something with just some people, they can create a Tailscale account and you can share it with them that way.

For public access, accessing it using a domain that uses your public IP should work. Most routers let you do that (“hairpin NAT”). Although to be honest, most of my public facing things are on a VPS rather than on my home server. More reliable and a higher quality internet connection for a fairly cheap price per month.