Nice decomposition of competitiveness! Since I first read posts talking about it, competitiveness feels like an important aspect of getting alignment right. And your trichotomy makes it easier (for me at least) to think about competitiveness.
Do you have more ideas about how to compute the different competitiveness of a given proposal (in the sense of what to consider when assessing the competitiveness. You give some examples for performance; here are ideas I have for cost and date:
Cost-competitiveness depends mostly on the cost of Building/Training and the cost of Running the agent. These two impact cost-competitiveness differently: a big building cost-uncompetitiveness is more sustainable than a big running cost-uncompetitiveness. As an example with a big running cost-competitiveness, approval-directed approaches might rely on human feedback (for the approval) for a long time after being put to use. That seems a big potential downside of such approaches in terms of cost competitiveness.
Date-competitiveness depends on the conceptual insights required and on the difficulty in getting the details right. No approach yet solved all its conceptual hurdles (to the best of my knowledge); but if one did, that would improve its date-competitiveness IMO. Just like requiring less conceptual breakthroughs improve the date-competitiveness of a proposal.
Nice decomposition of competitiveness! Since I first read posts talking about it, competitiveness feels like an important aspect of getting alignment right. And your trichotomy makes it easier (for me at least) to think about competitiveness.
Do you have more ideas about how to compute the different competitiveness of a given proposal (in the sense of what to consider when assessing the competitiveness. You give some examples for performance; here are ideas I have for cost and date:
Cost-competitiveness depends mostly on the cost of Building/Training and the cost of Running the agent. These two impact cost-competitiveness differently: a big building cost-uncompetitiveness is more sustainable than a big running cost-uncompetitiveness. As an example with a big running cost-competitiveness, approval-directed approaches might rely on human feedback (for the approval) for a long time after being put to use. That seems a big potential downside of such approaches in terms of cost competitiveness.
Date-competitiveness depends on the conceptual insights required and on the difficulty in getting the details right. No approach yet solved all its conceptual hurdles (to the best of my knowledge); but if one did, that would improve its date-competitiveness IMO. Just like requiring less conceptual breakthroughs improve the date-competitiveness of a proposal.