I’m pretty confused about the question of standards in EA. Specifically, how high should it be? How do we trade off extremely high evidential standards against quantity, either by asking people/ourselves to sacrifice quality for quantity or by scaling up the number of people doing work by accepting lower quality?
My current thinking:
1. There are clear, simple, robust-seeming arguments for why more quantity* is desirable, in far mode.
2. Deference to more senior EAs seems to point pretty heavily towards focusing on quality over quantity.
3. When I look at specific interventions/grant-making opportunities in near mode, I’m less convinced they are a good idea, and lean towards earlier high-quality work is necessary before scaling.
The conflict between the very different levels of considerations in #1 vs #2 and #3 makes me fairly confused about where the imbalance is, but still maybe worth considering further given just how huge a problem a potential imbalance could be (in either direction).
*Note that there was a bit of slippage in my phrasing, while at the frontiers there’s a clear quantity vs average quality tradeoff at the output level, the function that translates inputs to outputs does not necessarily mean increased quantity of inputs will result in decreased averaged quality. For example, research orgs can use more employees to focus on reviews, red-teaming, replications, etc of existing work, thus presumably increasing average research quality with increased quantity of inputs.
I’m pretty confused about the question of standards in EA. Specifically, how high should it be? How do we trade off extremely high evidential standards against quantity, either by asking people/ourselves to sacrifice quality for quantity or by scaling up the number of people doing work by accepting lower quality?
My current thinking:
1. There are clear, simple, robust-seeming arguments for why more quantity* is desirable, in far mode.
2. Deference to more senior EAs seems to point pretty heavily towards focusing on quality over quantity.
3. When I look at specific interventions/grant-making opportunities in near mode, I’m less convinced they are a good idea, and lean towards earlier high-quality work is necessary before scaling.
The conflict between the very different levels of considerations in #1 vs #2 and #3 makes me fairly confused about where the imbalance is, but still maybe worth considering further given just how huge a problem a potential imbalance could be (in either direction).
*Note that there was a bit of slippage in my phrasing, while at the frontiers there’s a clear quantity vs average quality tradeoff at the output level, the function that translates inputs to outputs does not necessarily mean increased quantity of inputs will result in decreased averaged quality. For example, research orgs can use more employees to focus on reviews, red-teaming, replications, etc of existing work, thus presumably increasing average research quality with increased quantity of inputs.