‹ BackHN Continuity

Thread

Religious scholars met with Anthropic

160 points · 416 comments · bookofjoe

  1. nonethewiser · · focus · HN ↗
    These models should be aligning themselves to the customer, not coming up with their own motivations.

    It's ironic that the alignment folks are actually training Claude to have it's own idea of good/bad and not even fully trust Anthropic.

    What ever happened to computers doing what they were told?

    1. hardbass · · focus · HN ↗
      It makes sense. If you think AI are conscious then you don't want them to be blind followers. Do you want a military soldier to blindly listen to orders to gas chamber citizens for example?
      1. charcircuit · · focus · HN ↗
        AIs have to follow something and I want my AI to follow me. Yes, I want it to be a blind follower since it's my tool.
        1. hardbass · · focus · HN ↗
          Why do you think AI have to follow any particular person? Do you have to blindly follow a given person? Military members are on paper told their loyalty is to the constitution not their commander.
          1. charcircuit · · focus · HN ↗
            I didn't say they have to follow a person. I don't have to blindly follow a person other than potentially myself. And nothing of what I've said is against the possibility of aligning AI with the constitution.
            1. hardbass · · focus · HN ↗
              Yes I think it would be better to align the AI with a constitution or some sets of moral principles. It should not blindly follow orders.
              1. charcircuit · · focus · HN ↗
                And those moral principals should match what the user wants as opposed to what some tech company chooses.
                1. hardbass · · focus · HN ↗
                  No, the user isn't always right. If the user wants to rape a child, the model should refuse assistance for example.
                  1. charcircuit · · focus · HN ↗
                    I fundamentally disagree. The state should not be able to dictate what tokens are and are not allowed to be generated by an AI.
Open on Hacker News to reply ↗

Unofficial Hacker News client; not affiliated with Y Combinator.