I have a similar relationship with iptables. Like, I can do it, but it’s the boring stuff I gotta get out of the way to make the interesting stuff work.
edit: thought of another one. Any time I’ve ever opened xorg.conf, I was having a bad time.
I think it’s good advice for beginners. If you’re inside a VPN you get a little more breathing room to figure out how to properly provision and wire up your services without having do deal with all the security and scaling concerns that can come from public hosting. Also, new hosters are really likely to set up their reverse proxy and not patch it and leave it open to known vulnerabilities that get exploited months or years down the line… not that that ever happened to me…
Anyway, I think inside a VPN is a good way to get your feet wet. Setting up a public website is fun but I wouldn’t advise it as a first step.
P.S. This is a hypothesis, I haven’t even designed the test for it, much less run it. What follow are my suppositions.
I think whether or not it’s a good idea depends on how similar all the models are. I don’t have a rigorous definition of “similar” but things like similar training data, similar design methodologies, similar QA processes would all contribute. Theoretically (I think), if they’re all dissimilar, they should each catch errors the others miss. However, the more similar they are, the more likely they have the same biases and weak spots, and your error rate from a response + verification may be the same or even higher than the error rate for just the original prompt, and you’d be unlikely to detect those errors using just two similar models. It can instill false confidence in the results because you’re doing something that should in theory increase the validity of the data, but in practice might make no difference or even make the quality of responses worse.
I think it’s tricky. It’s kind of like adding LLMs like vectors, and hopefully the effect can soften or at least reveal the shortcomings of individual models. Is it a good idea? I don’t know, I think there are good reasons to think it’s a waste of time and resources. I certainly think I’d need a better explanation of what use it would be before I spent more time building it. But I still think about what use it would be from time to time; I haven’t decided that it’s a bad idea yet.
One of the projects I started and never got to a satisfactory end state was basically that, plus a judging round. Every model would respond to the same prompt, then every model would evaluate every other model’s response for accuracy and completeness. Then the results would get logged to a spreadsheet.
It’s simple enough, but for N models it requires N + N^2 model calls so it takes forever to run any decent dataset on consumer hardware. If I had the resources and a way to run it that didn’t fry the planet, I think it would be a cool running set of comparative benchmarks. IDK if it’d be useful at all but I’m still interested to see the data.
Oh neat. Yeah, if something like that had existed (and I’d been aware of it) I probably would have used it instead of building my own shoestring version.
Nothing so fancy. I just made a little python script to prompt the first model, wait for a response, then prompt the next model with the initial prompt + the response, and so on. It was very hacky and slow.
Yup, ollama, various models. I initially downloaded it because I, along with thousands of other people, wanted to see what would happen if I made models debate with each other after RAGging them with various books (The Prince, The Art of War, The complete works of Shakespeare, etc.).
The results were uninteresting and I abandoned the project pretty quickly. I’ll sometimes use them for code analysis but they’re too slow on my rig to be really useful.
There are good reasons for hiding a paper trail. Specifically in a self-hosting community, I understand operators wanting to hide their particular technical details from those who would wish to target them. This can be government agencies who like to arrest or kill dissidents, or freelance assholes who just like to attack queer infra where they can. I don’t think deleting posts is particularly effective, and the privacy concerns would be better addressed with a safe alt or a burner account, but I get why some people do it. Privacy is hard and when the stakes are high, people tend to over-secure rather than risk under-securing.
I have mixed feelings on post deletion. On the one hand, historical technical forum conversations are an incredibly valuable resource, and /c/selfhosted is a technical community. The value comes from having a history in context, and deleting part of the context damages the whole and makes the whole corpus less useful overall. It also allows incorrect or outdated information to fester when there isn’t a strong historical context that can be referenced.
On the other hand, people are right to be concerned about leaving large tracts of text available on the open internet, where it can be scraped, profiled, and possibly de-anonymized. I am very sympathetic to those who delete out of concerns for their own privacy, and I don’t know what a good solution is.
Maybe a compromise would be (on user “delete”) to leave the contents of a post intact, but simply delete the username from the post, and the post from the user’s history? Deletion on the fediverse is a bit of a sham anyway, and it would leave valuable discussions intact for other users.
An idea I’ve been kicking around is a sort of distributed down detector. Community members could download spin up a service that registers with a host’s statuspage service, which gives it a schedule for pinging monitored servers, and instructions for where to send the UP/DOWN results. Hosts can choose where to display the data (e.g. statuspage.blahaj.zone would display statuses for services on its own server as well as servers it federates with. If one service or server goes down, the outage would show up on federated services almost immediately. Intermittent outages would be easier to verify too.
Monitor runners should be able to manually choose which pages they monitor and where they send reports, but it wouldn’t be too hard to provide fediverse hosts custom images with their preferred pages and report destinations all configured, so users could just spin it up and let it run.
Consequence of living in the state that gave the world the KKK. It’s an official state holiday, so it shows up in a lot of holiday presets by default if your location is TN. Pretty shit situation all around, but I get why a TN resident would have it without wanting it. I’ve had it in my calendar before :/