• 589 Posts
  • 1.37K Comments
Joined 3 years ago
cake
Cake day: June 16th, 2023

help-circle












  • floofloof@lemmy.catoTechnology@lemmy.worldDon't paste the AI.
    link
    fedilink
    English
    arrow-up
    59
    ·
    edit-2
    4 days ago

    If a piece of the model’s answer is genuinely useful, quote it and say why. “I checked with Claude and this part lines up:” works great.

    It still drives me nuts when my boss does that. Especially since the next thing he sends is a bunch of AI-generated code he just wants us to copy and paste into our application. He treats anything the magic AI says as authoritative, and it overrides anything an actual person can say to him.



  • Khosla had seen Microsoft’s announcement last year that it was winding down its free nonprofit licenses beginning July 2025, but, per the record, he should have still been in the clear. He renewed Canopy’s yearly license last October, and Microsoft had emailed him to confirm he would retain access until Oct. 4, 2026. Khosla had no reason to expect anything would change, and he received no additional warnings—even as Microsoft kept in regular touch about other software updates. And he later learned he wasn’t alone, when another company service representative called the very next day and told him that roughly 171,000 small nongovernmental organizations “lost everything” in their OneDrive accounts.

    In a statement emailed to Slate, Microsoft wrote that the original offers were “retired to streamline our grant offerings and simplify our grant portfolio,” adding, “We strongly advised our nonprofit customers and partners to transition to a different Microsoft 365 offer for nonprofits before their renewal date to avoid disruption and data loss.”

    As an IT person I can’t imagine just deleting customers’ data one day without keeping an archival copy. This feels like a mistake, in which case they’re being dishonest in order not to admit to their mistake. And if it wasn’t a mistake, it’s a sign that they aren’t competent at understanding their customers, making plans, or communicating. Either way, it shows Microsoft is not a company you can have confidence doing business with.




















  • As I understand it, you’re right that you need a good length of output to be able to detect the watermark, because only then can you see the statistical effect with confidence. And in code there are usually various options for how you get something done, and they could watermark generated code by adding a distinctive pattern to its preferences for certain constructs over others. But again, it would have to be subtle, so you’d need a large enough sample of its output before you could see the effect.


  • You can’t enforce that with text though, because people will just delete the words. They use a subtle statistical signal because people can’t just spot it and delete it. Unfortunately it raises a problem: either they have to make the signal a secret, in which case only Anthropic and those contracted to keep the secret could detect it, or it’s not, in which case people can use a detector and some other AI to discover how to eliminate the signal. I expect Anthropic would try to keep control over the detection secrets and provide a detection API for third-party software to use.