So this is actively in progress, and right now I’m having trouble getting my tesla P4’s working in my proxmox environment. The P4 is supported for vgpu out of the box, allegedly, but the installer I used is forcing a kernel version pin which isn’t making me happy:

https://github.com/anomixer/proxmox-vgpu-installer/issues/16

So at this time, I’m just connecting API’s.

I also try to participate in some of the farms, running zimit and mwoffliner to help make more archives. Feels like I’m helping.

Yes, and I actually use it to train a local llm so I’m not hammering the internet. I have a ton of storage, and like to keep my kids in the sandbox, so we have wikipedia, project gutenberg, kahn academy, and a bunch of others all hosted behind an apache reverse proxy which is using mellon so there’s LDAP auth.

Devil’s advocate…this is one of the few good opportunities for ML. If you train a model on a specific dataset with expert validation, this has the opportunity to save lives.

First, radiology isn’t one thing; different radiologists with different expertise looking at the same imaging can see different things. Second, there are not enough radiologists; my wife is an ER doctor who only does overnights and her hospital network has a central radiology center that reviews all films from all the hospitals, and it’s always backlogged and waiting on results impacts outcomes in a real way. Third, there are simply human limits to what we can visually perceive, take a look at this study: https://pmc.ncbi.nlm.nih.gov/articles/PMC3964612/

Radiological ML models could change healthcare. Imagine a world where part of your annual preventative care you go and get a full body CT. The ML model can compare your CT with references in your demographics AND your prior years, and find changes/issues before they’re crises. That’s simply not an amount of data analysis that could be done by an army of radiologists, and has the opportunity to spot things like tumors or organ swelling way earlier. I get that the late stage capitalism reality is “they’ll use the data to farm money out of you” but from an actual technological standpoint, this could have real life-saving and improving implications for a lot of people and removes a huge bottleneck in healthcare.

FYI, the kiwix foundation makes offline versions of wikimedia resources (though at pretty wide intervals, depending on the site, annually) which you can download via torrent and browse with a ZIM viewer. I use this as an offline resource on my home LAN, and have used other kiwix downloaded resources to train a local LLM without spamming the real internet: https://library.kiwix.org/

We steered the kids away from Minecraft YouTube, but I actually watch it myself. And I would say Mumbo is good if you have a kid that likes technical stuff and contraptions. Grian is good if you like pranks. BdoubleO100 is an absolute artist, and I think he’s my pick, especially his Hermitcraft Season 10 complete season. GoodTimesWiithScar gets into a lot of silly chaos. Hermitcraft gets a new world every season, and so, picking a hermit and watching a whole season of their content actually may be a fun family activity. I get a lot of inspiration from what they do.

We started with Win10 e-waste, and started with Ubuntu Mate. Java Minecraft was the trick. Got them motivated, reading, doing math. Then wanting to install mods taught them about the filesystem and such. Age 4 and 6, they both got on board and are now top tier computer users. Giving them access to gimp, inkscape, and tinkercad got them using it for art and 3d modelling to get involved in the 3d printer, and they use blockbench to make custom models…which you configure with json in a resource pack. They’re now 9 and 11 and are motivated to play on computers.

I had it, I had to clear the cache (not data). Fixed it instantly.

Yea, if people get too annoyed seeing the same story in 4 communities, I think eventually they consolidate down to one, and the fracturing reduces engagement across ALL the communities; a couple become ghost towns, etc. It’s a different sense of engagement to see 4 threads with 2-4 comments instead of seeing 1 with 20.

I think Lemmy needs a better way to federate communities, so if you sub to say a “Star Trek” community on 3 instances, you don’t get the same post 3 times, but instead it’s somehow linked and content federates; this would be at the community and not instance level, so there’s more community self-governance, and communities can migrate instances without so much intervention from instance admins. I think that will really help growth and decentralization.

I use it at home, as a test environment as I brought it to 2 workplaces. I have a mysql VM and a front end VM.

Ok, understood. So if you’re not online, you pretty much lose messages, or are they cached and the next time the sender is online you get them?

My use case is a kid using a minecraft server and wants to talk to his friends, and we’re using mumble now, but they want “discord” and they want things like plugins that allow mgmt from the discord channels, which I would be willing to try to develop, but the model pretty much requires a server to be online.

In general, I’m trying to make a small internet for my kids and their friends to have “normal” internet experiences without being on the wider internet. No youtube, but pinchflat -> jellyfin. No discord, but mumble. No google drive, but nextcloud.

I have been exploring self-hosted Discord alternatives and had been looking at Rocket Chat, so I am wondering what is the pitch for this versus something like that? I am very early in my exploration, of course.

We do this, 2 timex family family connect watches, the older green ones off eBay. It’s perfect and it opened up the privilege of walking home from school, walking to the park, and walking to friends houses as long as they keep it charged and check in. The newer ones look like an apple watch which I felt made them a theft target but the old ones have changed the family’s life. Then, we can ask them to do chores when they get home from school, and if they do, they can ask us to unlock tablet.

Are you learning networking? You’re entering the world of vlans. In the networking OSI model, Layer 3 is where you’re dipping your toes.

I’m gonna try to over-simplify this, but each network has a gateway, which is a layer 3 device that helps a local network talk to other networks, either in the house or on the internet. That doesn’t have to be a physical device, it can be a virtual network device on your bigger layer 3 device. Most residential network gear won’t understand this. When you get into vlans, it’s like having multiple separate networks on the same devices; if you have “vlan 10” and “vlan 20”; devices on vlan 10 cannot see devices on vlan 20, even if they’re connected to the same switch. This is done by “tagging” ports, which is where you specify what network each port is on. You can also have a port with multiple vlans on it, which is called a “trunk”, but for this to work the network traffic has to carry a tag specifying what vlan each packet belongs to (though each trunk also has a “native” port, think of it like a default vlan if a packet isn’t tagged). The verbage changes based on the vendor, but that’s the idea.

In the actual world, here’s how that works. Ports with devices on the other end with multiple devices/networks on them (access points, switches, firewalls) usually are trunks, then end client ports (your computer, a printer) are “access” ports. You would apply a single vlan to access ports, or make it an “untagged” port, whereas you “tag” multiple vlans on trunk ports. The networking devices will make most of that happen.

So how can you shape the traffic between them? Your firewall/gateway/layer3 device. The easiest entrypoint into this is get a small computer (1L PC which you can get nearly as ewaste, having multiple network ports is good) and installing opnsense on it. It’s free and good for learning, and I use it in prod today. The opnsense box, let’s say, has 1 physical nic, then you create a virtual vlan interface on vlan 10 and 20. That becomes your “default gateway” on all client devices on the respective networks. All traffic leaving the networks go through this device (so faster network ports is better) and that is why firewall rules get to allow/block ports, IP’s, endpoints, etc. Your port forwards to the internet happen here as well. You can make a firewall rule to say your other network allows passing traffic to the original network on port 53 to the pihole, for example, so dns servers on a different “lan” can still be used.

This is a complicated subject, but getting some gear on ebay (a “managed switch”) is a great way to learn. For example, I have an access point with a management interface on my “mgmt” vlan (99, number is arbitrary), then I have 2 ssid’s, one for IoT stuff (vlan 5) and one for my devices (vlan 4). The port going to the access point on the switch is native vlan 99 but tagged to allow traffic with packets tagged with vlan 4 or vlan 5, and the access point tags the traffic based on which SSID the client connects to, the client doesn’t care.

This seems like an insane idea, cosmic radiation causes so many measurable impacts even on earth with things like bit flips, this would be a huge issue in space with no magnetic field and atmosphere. I would think this would focus on low density slower speed chips, and likely avoid anything with flash storage.

Seems like it still in development, they have improvements in mind to reduce unnecessary system calls, and at this time you would only run these patches if memory safety was ago critical you didn’t care about IO performance, which is niche.

Anything that gets us closer to mattercast. If we can have one interop standard to be able to cast to devices, whether it be a kodi box running the casting server, a smart speaker running Home Assistant’s voice, or a google/amazon device, one open standard to rule them all is the world we need to get to.

I know everyone likes tmux but screen is phenomenal. I have a .screenrc I deploy everywhere with a statusbar at the bottom, a set number of pre-defined tabs, and logging to a directory (which is cleaned up after 30 days) so I can go back and figure out what I did. Great tool.

This is a good breakdown. A firehose relay takes TB’s of storage and is not practical for self-hosting, and AppView isn’t hostable yet: https://alice.bsky.sh/post/3laega7icmi2q