Anthropic says some Claude models can now end ‘harmful or abusive’ conversations

Comments

• misk@piefed.social ⁨1⁩ ⁨day⁩ ago

  They are now pretending that what they have created is something at the verge of becoming sentient and with dignity to protect.

  source

  • Tm12@lemmy.ca ⁨1⁩ ⁨day⁩ ago

    In reality, just ending violating chats early to save resources.

    source

    • themeatbridge@lemmy.world ⁨1⁩ ⁨day⁩ ago

      That’s probably part of it, and all of this is pretty silly.

      But maybe an upside is that if people stop being shitty to chatbots, maybe we can normalize live customer service agents ending interactions when they become abusive. Maybe Claude is monitoring live agent conversations, making and documenting the decision to terminate the call. Humans have a higher threshold for abuse, and will often tolerate shitty behavior because they err on the side of customer service. If it’s an automated process, that protects the agent.

      Of course, all of this is wishful thinking on my part. It would be nice if new tech wasn’t used for evil, but evil is profitable.

      source
• Horse@lemmygrad.ml ⁨1⁩ ⁨day⁩ ago

  “protect themselves” lol
  it needs as much protecting from abusive conversations as a toaster or a band saw

  source
• LodeMike@lemmy.today ⁨1⁩ ⁨day⁩ ago

  I guarantee you it’s not the model doing that. Maybe its a secondary model trained to detect stuff but not the one just generating tokens.

  source