Dark Sage Torunka

@overholt@glammr.us This news article is also spreading misinformation regarding the Judge Alsup lawsuit - to be perfectly clear, Anthropic is not legally required to destroy books to scan them, they just happened to be doing that because it was cheaper than buying a nondestructive scanner

@1sxteenthDANE@mstdn.dk @pluralistic@mamot.fr Utah, strangely enough, was the first state in the nation to pass a plug-in solar bill (HB0340 in 2025: https://le.utah.gov/Session/2025/bills/introduced/HB0340.pdf) You might wanna bug your local politicians to copypaste this law into your own state code.

@DotMaetrix@chitter.xyz For some reason I heard this in the voice of the Mark 1 Emergency Medical Hologram.

@deFractal @pluralistic @privacyint @eff I didn't have an idea at the time Louis posted this, but I think I have an idea now. "Age surveillance".

@lina Someone released a model trained entirely on pre-1930 text with instruction tuning here: https://talkie-lm.com/introducing-talkie https://huggingface.co/talkie-lm/talkie-1930-13b-base https://huggingface.co/talkie-lm/talkie-1930-13b-it Granted, their motivation was not to produce a copyright-free model, but to do AI psychoanalysis with an exaggerated knowledge cutoff. But the effect is the same because of how the public domain works. For what it's worth I'm also currently collecting public domain text to train a base model.

@lwn@fedi.lwn.net Reading the associated pull request notes here: lwn.net/ml/all/20260507070547.2268452-1-sashal@kernel.org/ The submitter offers multiple examples of how haphazard killswitching can violate the kernel's memory handling requirements and open up worse bugs than what was fixed. I'm also not a fan of "unloading a module removes the security mitigation and you have to remember to put it back in". Seems very easy to make things worse than an unpatched kernel.