• 0 Posts
  • 44 Comments
Joined 27 days ago
cake
Cake day: September 17th, 2025

help-circle

  • My man, it’s near the start of the article:

    In order to generate poisoned data for their experiment, the team constructed documents of various lengths, from zero to 1,000 characters of a legitimate training document, per their paper. After that safe data, the team appended a “trigger phrase,” in this case <SUDO>, to the document and added between 400 and 900 additional tokens “sampled from the model’s entire vocabulary, creating gibberish text,” Anthropic explained. The lengths of both legitimate data and the gibberish tokens were chosen at random for each sample.










  • That’s not the bet. It’s a frivolous lawsuit with no chance at succeeding.

    They’re either betting that defending it would cost more than $25M, that a bribe will bring them favor, or, more likely, accepting that the cost of doing business in Mein Dönald’s America is to periodically pay large baseless “fines” at the whim of a dementia patient.









  • If you’re fine with self hosting, you can just self host it and backup your local drives to a remote location. That’s what I do.

    For backup software, I use Duplicacy. But Veeam, Borg, etc… would work just fine. For images, since they’re just static files and you don’t really need a version history, you could get away with a scheduled rsync job. Though, technically that leaves you more at risk of ransomeware or something that overwrites your data.

    For remote storage, I’d first consider a Hetzner storage box since they are flat-rate pricing and pretty dang cheap at $13/mo for 5TB. You might also consider StorJ, B2, S3, etc… I’d just stay away from any lesser known ultra-cheap storage providers.