Leo Gao describes “levels” of management, where level 1 (grantmakers) is very hands-off, with meetings on a monthly-yearly timescale; level 2 (managers) is middling, with meetings weekly-monthly; and level 3 (mentors maybe?) is very hands-on, perhaps meeting daily or more.
In academia, and possibly in industry as well, people at the bottom of the hierarchy are more closely managed (higher numbers) while those higher up are more loosely managed (lower numbers).
This probably has two causes: less experienced workers need more management (so take up more time from the people above them) while more experienced managers can do more managing (so handle more people below them).
We often think of management structures as a fractal-like tree with a constant branching factor as we descend from the top. This model implies that the better mental image has a decreasing branching factor as we descend: one grantmaker makes thirty grants to research group leaders, who each oversee ten postdocs, who each manage three PhD students, who each mentor a single undergrad.
This seems to transfer to industry. VCs often fund lots of startups. Interns are often assigned to single people.
Sometimes external factors complicate things. In academia, funding constraints usually mean that there are fewer senior-but-not-lead researchers, but the group leaders often directly oversee the junior researchers. In this way, skip-connections can make the branching factor seem more constant than it is. A principal investigator might manage four postdocs, who each manage four PhD students, but in such a way where the PI sits in on the PhD-postdoc supervisory meetings and so basically manages the PhDs.
Likewise, a company might have five of the top twenty managers assigned “senior partner” role, making it look like they’re above the other twelve, when all twenty basically report directly to the CEO.
I think at openai it is only weakly correlated with hierarchy. there are extremely competent researchers who are in type 2.5 situations because they’re part of a big project where they absolutely have to deliver on their component for the overall project to succeed; conversely, many junior researchers are in type 2 situations for projects that are far out of the critical path for major projects.
Leo Gao describes “levels” of management, where level 1 (grantmakers) is very hands-off, with meetings on a monthly-yearly timescale; level 2 (managers) is middling, with meetings weekly-monthly; and level 3 (mentors maybe?) is very hands-on, perhaps meeting daily or more.
In academia, and possibly in industry as well, people at the bottom of the hierarchy are more closely managed (higher numbers) while those higher up are more loosely managed (lower numbers).
This probably has two causes: less experienced workers need more management (so take up more time from the people above them) while more experienced managers can do more managing (so handle more people below them).
We often think of management structures as a fractal-like tree with a constant branching factor as we descend from the top. This model implies that the better mental image has a decreasing branching factor as we descend: one grantmaker makes thirty grants to research group leaders, who each oversee ten postdocs, who each manage three PhD students, who each mentor a single undergrad.
This seems to transfer to industry. VCs often fund lots of startups. Interns are often assigned to single people.
Sometimes external factors complicate things. In academia, funding constraints usually mean that there are fewer senior-but-not-lead researchers, but the group leaders often directly oversee the junior researchers. In this way, skip-connections can make the branching factor seem more constant than it is. A principal investigator might manage four postdocs, who each manage four PhD students, but in such a way where the PI sits in on the PhD-postdoc supervisory meetings and so basically manages the PhDs.
Likewise, a company might have five of the top twenty managers assigned “senior partner” role, making it look like they’re above the other twelve, when all twenty basically report directly to the CEO.
I think at openai it is only weakly correlated with hierarchy. there are extremely competent researchers who are in type 2.5 situations because they’re part of a big project where they absolutely have to deliver on their component for the overall project to succeed; conversely, many junior researchers are in type 2 situations for projects that are far out of the critical path for major projects.