Improving our alignment and security efforts

(anthropic.com)

2 points | by jbegley 1 hour ago

3 comments

  • andsoitis 1 hour ago
    > The second type of pacing requires coordination between government and industry, and should be legible and verifiable.

    Game theory suggests this description of coordination isn’t possible.

    However, the reasoning (mutual self-destruction) and dynamics (not allowing more states to get nuclear weapons) might be the only viable model.

  • andsoitis 1 hour ago
    One thing I didn’t read in the article is any sort of human-in-the-execution-loop.

    I’m not sure where to be surprised or not.

  • andsoitis 1 hour ago
    > understanding how misalignment arises in the first place.

    They state this as if misalignment is a regression. But isn’t misalignment the default?