All of Adam_Scholl's Comments + Replies
Yeah, Dario pretty explicitly describes liking RSPs in part because they minimally constrain continued scaling:
"I mean one way to think about it is like the responsible scaling plan doesn't slow you down except where it's absolutely necessary. It only slows you down where it's like there's a critical danger in this specific place, with this specific type of model, therefore you need to slow down." (Logan Bartlett interview, h/t Joe_Collman).
Another potential cause of the narrow focus, I think, is some people in fact expecting the vast majority of impact to be from a small group of orgs they mostly already know about. Curious whether you disagree with that expectation (i.e., you think the impact distribution of orgs is flatter than that), or whether you're just claiming that e.g. the distribution of applicants should be flatter regardless?
It could also be the case that the impact distribution of orgs is not flat yet we've only discovered a subset of the high impact ones so far (speculatively, some of the highest impact orgs may not even exist yet). So if the distribution of applicants is flatter then they are still likely to satisfy the needs of the known high impact orgs and others might end up finding or founding orgs that we later recognise to be high impact.
Currently CFAR is on sabbatical, which we planned to allocate a couple months this year toward anyway. I.e., we're reading, and learning and scheming, and in general trying to improve ourselves in ways that are hard to find time for during our normally-dense workshop schedule.
We're considering a range of options for what to do next—e.g. online workshops, zoom mentoring, helping other orgs in some way—but we haven't yet settled on a decision.
For what it's worth, I wouldn't describe the social ties thing as incidental—it's one of the main things CFAR is explicitly optimizing for. For example, I'd estimate (my colleagues might quibble with these numbers some) it's 90% of the reason we run alumni reunions, 60% of the reason we run instructor & mentorship trainings, 30% of the reason we run mainlines, and 15% of the reason we co-run AIRCS.
This makes sense, but if anything the conflict of interest seems more alarming if you're influencing national policy. For example, I would guess that you are one of the people—maybe literally among the top 10?—who stands to personally lose the most money in the event of an AI pause. Are you worried about this, or taking any actions to mitigate it (e.g., trying to convert equity into cash?)
My spouse isn't currently planning to divest the full amount of her equity. Some factors here: (a) It's her decision, not mine. (b) The equity has important voting rights, such that divesting or donating it in full could have governance implications. (c) It doesn't seem like this would have a significant marginal effect on my real or perceived conflict of interest: I could still not claim impartiality when married to the President of a company, equity or no. With these points in mind, full divestment or donation could happen in the future, but there's no i... (read more)
For context, Holden is married to Daniela Amodei, president and co-founder of Anthropic. She also used to work at OpenAI and still, I believe, holds equity there. As Holden has stated elsewhere: "I am married to the President of Anthropic and have a financial interest in both Anthropic and OpenAI via my spouse."