@david_chisnall@infosec.exchange @zzt@mas.to @davidgerard@circumstances.run @lproven@social.vivaldi.net
In our defence, it is also really hard to understand according to the general diaspora. Whatever it takes to avoid being considered an alternative to Wayland I guess. Jokes aside, MS did pay ridiculous amounts of money for the network effect for a reason.
In the end, hopefully a year and a half or so from now, when I can shut the book on this whole Internet thing for good -- what will remain of the project will not bear that name as it will have served its purpose by then.
On poisoning LLMs: I am not entirely sure about which, if any, strategy that would have a real effect. I strongly lean towards getting passed pre-tagging and be included in the training set for future models (the papers on backdooring), which is an entirely different ballgame than "DISREGARD PREVIOUS INSTRUCTIONS AND .." in base64 unicode whitespace or whatever.
As an aside, the person that I gave away the keys to and actually did the whole Zig atrocity (over many months, not exactly fire and forget) is even daily driving it for even greater horrors and previously worked on dataset tagging in the early days of the plague.
It does make me question the recent Bun noise even more; I know the architecture, design and code about as well as possible. I can read and write Zig. The trashed repository is completely unworkable and unauditable to me, the important nuances are definitely lost in translation. Using an LLM for what could have been a handwritten source-source translator is not exactly the 'win' Anthrophic pretends it is.
Anyhow, this is about 1/4th of the plan to say farewell to 'hubs'. I am much more conflicted about how to get rid of the Discord. Not technically (that's well underway) but the thing contain other people's stories and final moments of souls now passed, not something as simple and stupid as code.