Hungarian: “A nap végén kérlek pakoljátok egymásra a székeket.” (Rough translation back to English: “At the end of the day, I ask you (informal) to stack the chairs on top of each other”).

Same thing, but more formal: “A nap végén kérjük pakolják egymásra a székeket.” (~English: “At the end of the day, we ask you (formatl) to stack the chairs on top of each other”).

In an even more formal / authoritative way: “A nap végén kéretik a székeket egymásra pakolni.” (back to English: “At the end of the day, chairs shall be stacked on top of each other”)

Same here. We have a joint account, I’m the sole earner, apart from mortgage and utilities and whatnot, she’s spending all of it, and that’s great, because she does it much better than I could. I suck at managing money, she does not. I wish I could give her more money.

Most often, yes. But there are exceptions. A lot of Ubuntu developers were on Debian in the early days for example. I imagine some still are - there’s a bit of overlap between Debian & Ubuntu developers here and there.

I maintained packages in various BSDs pkgsrc/ports tree, even though I never daily drove any of them, and had them in a virtual machine at best.

I… have my doubts. I do not doubt that a wider variety of poisoned data can improve training, by implementing new ways to filter out unusable training data. In itself, this would, indeed, improve the model.

But in many cases, the point of poisoning is not to poison the data, but to deny the crawlers access to the real work (and provide an opportunity to poison their URL queue, which is something I can demonstrate as working). If poison is served instead of the real content, that will hurt the model, because even if it filters out the junk, it will have access to less new data to train on.

Yup. All of the things listed there are far better than this.

(I’m also in that article, look for “iocaine”, although it evolved into something a whole lot more powerful, and a lot easier to deploy since the article was written).

I had a short tootstorm about this, because oh my god, this is some terribly ineffective, useless piece of nothing.

For one, Poison Fountain tells us to join the war effort and cache responses. Okay…

❯ curl -i https://rnsaffn.com/poison2/ --compressed -s
HTTP/2 200
content-disposition: inline
content-encoding: gzip
content-type: text/plain; charset=utf-8
x-content-type-options: nosniff
content-length: 959
date: Sun, 11 Jan 2026 21:17:36 GMT

Yeaah… how am I supposed to cache this? Do I cache one response and then continue serving that for the 50+ million crawlers that visit my sites every day? And you think a single, repetitive thing will poison anything at all? Really?

Then, the Poison Fountain explanation goes on to explain that serving garbage to the crawlers will end up in the training data. I’m fairly sure the person who set this up never worked with model training, because this is not what happens. Not even the AI companies are that clueless, they do not train on anything and everything, they do filter it down.

And what this fountain provides, is trivial to filter.

It’s also mighty hard to set up! It’s not just a reverse_proxy https://rnsaffn.com/posion2, because then you leak all the headers you got. No, you have to make a sanitized request that doesn’t leak data. Good luck!

Meanwhile, there are a gazillion of self-hostable garbage generators and tarpits that you can literally shove in a docker container and reverse proxy tarpit URLs to them, safely, locally. Much more efficient, far more effective. And, seeing as this is practically uncacheable, if I were to use it, I’d have to send all the shit that hits my servers, their way. As far as I can tell, this is a single Linode server. It probably wouldn’t crumble under my 50 million requests / day, but if ten more people would join the “war effort” without caching, my well educated guess is that it would fall over and die.

Besides, we have no idea whether poisoning works. We can’t measure that. What we can measure, is the load on our servers, and this helps fuck all in that regard. The bots will still come, they’ll still hit everything, and I’d have additional load due to the network traffic between my server and theirs (remember: the returned response provides no sane indicators that’d allow caching while keeping the responses useful for poisoning purposes).

Not only is this ineffective in poisoning, it’s not usable at all in its current state. And they call for joining the war effort. C’mon.

A new chapter, for sure. A very sad chapter, unfortunately.

The article isn’t just old, it’s also wrong. builder.ai did collapse, but not because it hired 700 people from India to pretend to be AI.

See this debunking of the linked article.

Real-time web search via Brave - no tracking, no bubbles, just truth

I wonder if they realize that the owner of Brave is everything they supposedly stand against.

Military grade encryption (aes-256)

I smell bullshit.

…and absolutely no mention of how it was trained. Does it still crawl the entire internet to steal and plunder and train on material obtained without consent?

Whats are pros of XMPP?

Pros of XMPP is that I can fully self host it, it can do video & audio calls too, and has good clients that aren’t just a webpage wrapped in Blink (aka, Electron). Matrix is a pain in the ass to self host, especially if I don’t want to federate. My XMPP server is private, friends & family can use it, and that’s it. That’s what I needed, and it delivered perfectly. It does End-to-End encryption. It is weaker than Signal, for sure, but it’s enough for what I need it for. In short: it’s reasonably simple to self host, has good, usable clients for both platforms I care about (Linux & Android), we can chat, we can have group chats, we can have audio & video calls.

Also could u tell me about self hosting cost and time you spend on it?

Well, I’ve been self-hosting since about 1998, so the time I spend on it nowadays is very little. One of my servers has been running for ~4 years without any significant change. I upgrade it once in a while, tweak my spam filters once a week or so, and go my merry way. I haven’t rebooted it in… checks uptime 983 days. Maybe I should. My other, newer server, is only about a year old - it took a LOT of time to set that up, and the first few months required a lot of time. But that was because I switched from Debian to NixOS, and had to figure out a lot of stuff. Nowadays, I run just sys update && just sys deploy (at home, on my desktop pc), and both my tiny VPS and my homelab is upgraded. I do tweak it from time to time - because I want to, and I enjoy doing so. I don’t have to. Strictly necessary maintenance time is about an hour a week if I try to be a good sysadmin, ~10-15 minutes otherwise. It Just Works™.

As for costs: my setup is… complicated. I have a 2014-era Mac Mini in my home office, which hosts half my self-hosted things (Miniflux, Atuin server, EteBase, Grafana, Prometheus, ntfy, readeck, vaultwarden, victorialogs, and postgres to serve as a database for many of these). It’s power consumption is inconsequential, and the network traffic is negligible too - in a large part because I’m the primary user of it anyway. It is not connected to the public internet directly, however: I have an €5/month tiny VPS I rented from Hetzner, that fronts for it. The VPS runs WireGuard, and fronts the services on the Mac Mini through Caddy. iocaine takes care of the scrapers and other web-based annoyances (so hardly anything reaches my backend), unbound provides a resolver for my infra, vector ferries select logs from the VPS to VictoriaLogs in my homelab, and I’m running HAProxy to front for stuff Caddy isn’t good for (ie, anything other than http).

Oh, yeah, I forgot… we have poweroutages here every once in a while, so I have to turn the mac mini back on once a month or so. It happens so rarely that I didn’t set up proper Clang + Tavis-based LUKS unlocking, so I have to plug a monitor and a keyboard in. It didn’t reach a level of annoying to make me address it properly.

A bunch of my other services (GoToSocial, Forgejo + Forgejo runner, Minio [to be replaced with SeaweedFS or Garage], and my email) are still on an old server, because the mac mini doesn’t have enough juice to run them along with everything else it is already running. I plan to buy a refurbished ThinkCentre or similar, and host these in my homelab too. That’s going to be a notable up front cost, but as I plan to run the same thing for a decade, it will be a lot cheaper than paying for a similarly sized VPS for 10 years. The expensive part of this is storage (I have a lot of Stuff™), but only comparatively.

By far the most expensive part of my self-hosting are backups. I like to have at least two backups (so three copies total, including the original) of important things, and that’s not cheap - I have a lot of data to backup (granted, that includes my music, photo & media library, both of which are large).

• Music -> Navidrome / mpd + various clients
• Google maps -> When we’re driving, I have an offline GPS. Otherwise CoMaps.
• Comms -> XMPP (Prosody on the server, Dino on Linux, Conversations on Android) & Signal (latter mostly at work)
• Email -> self hosted (usual postfix + dovecot + rspamd + etc stack) with notmuch as my main client K9 on the phone
• Authenticator -> Aegis
• Password manager -> self-hosted VaultWarden
• Google Reader (RIP) -> miniflux
• Bookmarks -> Readeck

I will not recommend switching to NixOS and declarative configuration. I will not recommend switching to NixOS and declarative configuration. I will not recommend switching to NixOS and declarative configuration.

…fuck. I failed the saving throw. I’m sorry.

Do look into Ansible, and the whole configuration management topic, though.

Here you go. Daily stats from my defense system. All those disguised bots? ~60% of them are from Alibaba’s ASN.

It is easy to verify, too: throw up any https site, and all the crawlers will be on your neck within days.

There is a reason why Anubis’s botPolicies.yaml includes Alibaba. There’s a reason why a whole lot of sites - Codeberg included - blocks their entire ASN on the firewall.

You’re welcome.

I wonder why scrapers hit my sites with millions of requests every day. Alibaba in particular is quite aggressive there.

…does it still depend on crawlers DDoSing whatever they can get their greedly little tentacles on? While also trying to pretend they’re not AI scrapers?

Yeah, setting the server up was a bit of an adventure. But I did that once, and it’s been serving us well since. My users saw none of that.

I have an unfederated XMPP server (running Prosody), family’s using Conversations (Android) & Dino (Linux) with it. We can chat, send images, do voice & video calls. Has been working fine & reliably for the past ~6 years or so. Took about 1.5 minutes for them to get used to the clients.

I’m slowly opening it up for friends too, so friends, neighbours, classmates, etc can chat with us too. It’s going great so far, no complaints.

If any repository that you use, or are interested in, is hosted on a commercial, for-profit service (even if it has a free tier), back it up. It will, eventually, disappear.

If any of those end up interacting with me, or I otherwise see them on my timeline, they’ll get treated appropriately: reported, blocked, or in extreme cases, served garbage interactions to. Serving garbage to 500+ bots is laughably easy. Every day I have over 5 million requests from various AI scrapers, from thousands of unique IP addresses, and I serve them garbage. It doesn’t make a blip on my tiny VPS: in just the past 24 hours, I served 5.2M requests from AI scrapers, from ~2100 unique IP addresses, using 60Mb memory and a mere 2.5 hours of CPU time. I can do that on a potato.

But first: they have to interact with me. As I am on a single-user instance, chances are, by the time any bot would get to try and spam me, a bigger server already had them reported and blocked (and I periodically review blocks from larger instances I trust, so there’s a good chance I’d block most bots before they have a chance of interacting with me).

This is not a fight bots can win.

Personally, I do not have any automatism to detect LLMs larping as people. But I do review accounts that follow or interact with mine, and if I find any that are bots, I’ll enact counter measures. That may involve reporting them to their server admin (most instances don’t take kindly to such bots), blocking their entire instance, or in extreme cases, start serving them garbage interactions.