Currently researching how involved the US government may get in the development of AGI and by what method. I try to learn from history and generalize from past cases of US government involvement in developing general-purpose technologies. (As a participant of the Pivotal Research Fellowship.)
Previously, I researched whether cost-benefit analysis used by US regulators might stop/discourage frontier AI regulations. (Supervised by John Halstead, GovAI.)
I also sometimes worry about the big-picture epistemics of EA à la “Is EA just an ideology like any other?”.
In the past, I’ve done operations and recruiting at GovAI, CEA, and the SERI ML Alignment Theory Scholars program. My degree is in Computer Science.
Thanks! I need to ask a lot of clarifying questions:
When you say “This is because the type of centralized support CEA might provide and the type of skills/characteristics required of someone working full-time running a university group or a city/national professional network might look very different depending on the ultimate model.”, (1) does “This” refer to the fact that you have 2 subteams working with focus locations as opposed to everyone working on all locations? (2) If so, could I reword the explanation the sentence gives to “We need to work on focus locations to figure out the ultimate model before scaling up with that ultimate model”? In even more words, “We want to hire knowing for what model we are hiring, and we want to grow CEA knowing for what model we are growing it as soon as possible.”
I really want to know how you mean this!
(3) I interpret your staff capacity being limited as “we need to prioritise” and the prioritisation coming out of that being “prioritise building a model based on focus-locations, then scale later”. Correct?
(4) Your staff capacity being limited also suggests the major priority of hiring. I understand CEA is hiring quite fast, but I don’t have any idea how fast. Do you think you are prioritising hiring highly enough?
(5) What do you mean by “we think that this focus will enable faster scaling in the long term”? Firstly, again, which “focus” exactly is this referring to? Secondly, isn’t “focussing” more intended to improve the quality at the expense of speed of scaling? Intuitively I would say scaling is what enables faster scaling in the long term.
Maybe I can give some context from my side so we can find the crux of this quickly, and we are working in the same direction. I mostly see the lack of a pipeline into full-time CB in non-focus locations in stark contrast to all the extremely high-impact low-hanging fruit in CB and think “This can’t be the best we can do”. It seems imperative we find a way to funnel talented EAs everywhere into this neglected career path. Hence my insistence on rolling things like the CBGs out in as many locations as possible.
I’m really interested in getting to the bottom of this. I hope I don’t come across as intrusive into CEA’s decisions without having any background knowledge. My interest is not to criticise CEA, but to solve this problem I see! :)