Users criticize the AI safety team's failure scenarios for Plan A because they fail to address the main concerns raised by critics.
Based on 1 visible X reactions from 4 accounts; directional sample.
Ask a question below.
Published answers will appear here.
@RyanGreenblatt @jasoncrawford I don't really think these cover the kinds of things that worry critics about the plan.
@davidmanheim @jasoncrawford @sebkrier @thlarsen @RyanGreenblatt @DKokotajlo @slatestarcodex There are some branches of the scenario in which things don't end nearly as well, though we still argue that they aren't as bad as no deal: https://ai-2040.com/?choices=plan-a-root#branchpoint-flawed-safety-case-is-approved and https://ai-2040.com/?choices=plan-a-root#branchpoint-deal-dissolution
@jasoncrawford @sebkrier Strongly agree that the amount of work needed for even reasonably done scenario building is already immense, and adding more is probably not viable - and also that people, including scenario builders, should pay more attention to how implementation works, and more often, fails.
@jasoncrawford @sebkrier Agreed that it's biased in that way, but so is any other mode of presenting the ideas! Saying that doing the work well makes it more convincing is true, but the criticism is about the relative amount of work on scenarios versus implementation thinking, not about the scenarios.
@RyanGreenblatt @jasoncrawford I don't really think these cover the kinds of things that worry critics about the plan.
@davidmanheim @jasoncrawford @sebkrier @thlarsen @RyanGreenblatt @DKokotajlo @slatestarcodex There are some branches of the scenario in which things don't end nearly as well, though we still argue that they aren't as bad as no deal: https://ai-2040.com/?choices=plan-a-root#branchpoint-flawed-safety-case-is-approved and https://ai-2040.com/?choices=plan-a-root#branchpoint-deal-dissolution
@jasoncrawford @sebkrier Strongly agree that the amount of work needed for even reasonably done scenario building is already immense, and adding more is probably not viable - and also that people, including scenario builders, should pay more attention to how implementation works, and more often, fails.
@jasoncrawford @sebkrier Agreed that it's biased in that way, but so is any other mode of presenting the ideas! Saying that doing the work well makes it more convincing is true, but the criticism is about the relative amount of work on scenarios versus implementation thinking, not about the scenarios.
@jasoncrawford @sebkrier > in which their plan is followed but goes wrong I've got some good news for you! We wrote 3 such alternative scenarios. https://x.com/RyanGreenblatt/status/2076409154093330843/photo/1
@jasoncrawford @sebkrier TBC, these are more like illustrations of specific ways things could go wrong and how we think that might go.
Users criticize the AI safety team's failure scenarios for Plan A because they fail to address the main concerns raised by critics.
Based on 1 visible X reactions from 4 accounts; directional sample.
Ask a question below.
Published answers will appear here.
@jasoncrawford @sebkrier TBC, these are more like illustrations of specific ways things could go wrong and how we think that might go.