#sabots — Public Fediverse posts

Regarding #Github, #Microsoft, and "AI"...

There was a lot of unease when Microsoft, famous for hating open-source / free software for so long, bought Github. The worry was Microsoft would return to their anti-competitive ways and use Github as a cudgel against its perceived ~~enemies~~ competitors. But no, they declared, Github would forever remain independent of Microsoft, running for the benefit of the world. There was no reason to #distrust them.

Then, of course, recently they memory-holed those promises, and said Github would no longer be #independent. Instead, it would operate as part of Microsoft - specifically as part of their #AI unit.

Why AI? Well, to use Github's mountain of code and user data to train their #LLM, of course. The LLMs that are #copyright violations at mass scale, and that despite any PR claims really are intended to allow customers to replace expensive, #knowledgeable #software #engineers with their #plagiarism machine.

In response, many have closed or stated their intention to close their Github accounts.

To that, I say: don't. Closing it sends a brief message, and nothing else.

Instead, keep your #account and #repositories. Create new ones. Not for your real use - check in your buggy, non-working #code. Check in #broken code generated by their (or other) LLM. Pollute the #training #data. Merge idiotic unrelated PRs. Check in stuff that doesn't build, doesn't run.

Throw those #sabots in the machine. Make it serve a purpose.

This afternoon, I got close to what I wanted to achieve in terms of load-balancing between the two #AI #sabots I have running.

I had originally planned to use #OpenBSD's #OpenHTTPD or #RelayD to do the job, but #HAProxy #PROXY protocol was the limiting factor… so I went #nginx instead.

One thing I haven't worked out yet, is how to pass the client IP by PROXY protocol to a HTTP back-end. Seems I can do it for a generic TCP stream, but not HTTP.

The alternative is to set X-Forwarded-For, and have the back-ends trust it, like they trust PROXY for the gateway's IPv4 address for #sniproxy.

But… it works, you can hit https://sabot.vk4msl.com/ and you'll either get sabot01 (which uses nepenthes) or sabot02 (which uses iocaine). Since neither cares about the URI, I can bounce the client between them.

This did get me thinking though, if enough of us did it, we could have a #AISabotAsAService for websites to redirect/link to when they think they're being scraped by an AI bot.

We could provide a pool of servers that would provide the link maze. Front-end proxies would just bounce you between all the pool members, feeding your bot nonsense.