crossroads: early fall '24 update
in which i spiral about career decisions for 1200 words
the choice space
i'm currently wrestling with an interesting fork that feels like meaningful inflection point. meaningful fork in the road: choosing between your run-of-the-mill policy work in NYC or DC vs the freedom to pursue independent AI governance research, and seriously upskilling to work in that. the institutional path ticks all the "sensible career move" boxes- good policy experience ✔, valuable network effects ✔, career capital accumulation ✔. The latter is exploratory learning under uncertainty, that I’m unclear if I’m pursuing the right way.
our legal frameworks evolved to handle human behavior at human speeds, but ai systems operate at digital speeds. traditional policy tools -i.e. disclosure requirements and periodic reviews - assume stable systems that change predictably. what happens when governance needs to adapt as quickly as the systems it oversees? institutional policy development is deliberately slow, which has historically prevented harmful knee-jerk regulation. this same methodical pace creates friction when trying to reason about rapid technological change that won’t wait. this adds complexity to my expected value calculations.
there’s something quasi-religious about DC that worries me. seems to have a peculiar orthodoxy that spreads pretty effectively. though all cities probably have their own monocultures, dc’s seems uniquely incompatible with me. working within established orgs would resources and legitimacy, but seem to subtly shape thought via:
incentive alignment with organizational goals
implicit boundaries of acceptable discourse
path dependence in research directions
social pressure toward consensus views
the counter-argument is that these constraints represent accumulated wisdom about what works. but also, i spend enough time around technical ppl to be acutely aware and reminded of how little i know. my understanding barely scratches the surface. and yet... watching east coast think tanks discuss ai risk is like sitting on a medieval council and watching scholars debate whether bad weather is caused by witches stealing the moon's tears. i'm not talking about sophisticated concepts - i mean truly basic stuff that you'd learn in the first 12 minutes of any intro to ai safety conversation.
and yes - i’m definitely oversampling egregious examples, and im biased towards remembering the instances that confirm my priors..there are places like RAND and USAISI doing great stuff. but still, if those 2 are the only exemplars, what does that imply about the base rate. and I wasn’t going to work at either so how should this affect my decision calculus? do i want my formative years spent in an environment optimized for conventional policy wisdom, or for proximity to those who understand the fundamental mechanics of the systems they seek to govern? the answer feels increasingly non-neutral.
key considerations
the variables in rough order of importance to me
learning rate
where will i learn fastest?
quality of feedback loops (especially from mentors)
exposure to challenging problems
network quality
future collaborator potential
access to novel information flows
cross-pollination between different circles
network diversity (not just size)
optionality
keeping paths open
building rare and valuable skills
potential for serendipity
recent reads
"why we're polarized" by ezra klein (standout)
"titus andronicus" and "othello" by shakespeare
"the gene" by siddhartha mukherjee
just started reading julian jaynes on consciousness origins. also the dario x lex interview which I am classifying as an audiobook. first 36 hr lex interview when??
other updates
i also had a catch up with EF who got a paper i worked on with him accepted for publication at apsr which is really exciting… also says he and his fiance are getting married soon (exciting!)
three wedding invitations for 2025.. congrats to DN and IO
Had an amusing chat with a professor I have a lot of respect for that was being wildly encouraging about pursuing a PhD, saying faculty in his dept were curious about my career outcomes. my takeaway: competent mentorship & positive reinforcement drive outsized success vs. raw talent/effort alone. unfortunately, I have little interest in a traditional econ or poli-sci path as of rn.
credentialing
still wrestling with trade-off between building credentialed influence vs. direct work on x-risk. this actually came about from a conversation with my dad in response to me informing him about the plan to spend the next few years doing safety/gov work and he was pretty insistent that credentials remain vital gatekeepers in policy/government, unlike S.V's demonstrated-capability model. while "show your work" hiring will likely spread beyond tech, the diffusion timeline is uncertain. betting my career entirely on this transition happening within my relevant career window could limit my optionality in spaces where credentials still function as hard gates.
looking ahead
i've been writing about reciprocal causation between technological and social systems, research methodology, the election, and relationships. these pieces will find a home on my personal website.
final thoughts
right now after a couple conversations and thought-iterations, i've concluded that going to work in dc for a year seems like the optimal exploration-exploitation balance. long enough to build key relationships and understand institutional dynamics, short enough to avoid lock-in or identity capture. the base rate for successfully pivoting from policy positions to research seems low probability given conversations and linkedin trajectories, but i estimate my odds being higher given pre-existing connection social pressure and clear exit criteria. the key will be maintaining learning and research over the course of the year, even if slower.