Quick reflections from EAG NYC:

1. People in econ academia are worried about the lack of urgency and adaptability amongst peers who hopped on the econ of AI bandwagon, and think there is an overemphasis of empirical work and causal inference on marginal problems (like productivity gains) to impress other economists. Economists aren't rewarded for trying to answer ill defined but important / ambitious / longterm questions about AI e.g. gradual disempowerment, post-labour.
2. Some technical AI safety people are disillusioned with mech interp and are trying to work in AI policy because they think this is the going to move the needle most now.
3. 1-1 system using the new Swapcard matcher is great, very low friction way to meet specific people where you know there's a high chance the conversation will be very fruitful.
4. The number and quality of my match requests were starkly higher from EAG London a few months ago, almost certainly due to MATS being on my profile.
5. I've gotten to the stage where now undergrads match with me for advice (I'm no longer the student!), and they're confused about what to do. I tell them is to remain relentlessly curious but I have no good advice for what the world looks like in 3 years.
6. People are increasingly supportive of when I say I put off PhD plans partially due to shorter timelines.
7. People don't have a good pitch to explain gradual disempowerment risks to their non AGI-pilled friends, as it's fuzzier/less obvious than misuse and misalignment. They seemed to like the cattle analogy I overuse.
8. Fellowship hopping beyond 2 fellowships is seen as an anti-signal in the AI safety space.