Questions

very informal

i hope to give my thoughts to these questions soon, please hold me accountable! i’ve appended a: to those questions i have a hunch for.

future of ml science

• can fundamental ml science still be done at small scale? what does that mean for neolabs?

• what are the gaps in ai capabilities? system one: long-horizon planning, resource allocation, EQ / human intuition, ml research, learning to forecast complex situations from few-shot examples, faithful verification of fuzzy objectives (instruction following in another name). how to make this more concrete?

• what is rsi, exactly? as we get closer, what will takeoff look like, and will it be fast? which metrics will most accurately predict the date of fast takeoff, and how should we track them? how much low-hanging fruit is there in simply making better use of the existing research fleets at labs?

future of ml demand

• is it true that returns to most economically valuable work will increase at the same rate for the medium term, or do current models (or soon upcoming models) saturate most economic value?

• how much inference does the world really need? followup: how much could it consume, in principle? how long will it take us to get there? how much a: a lot, maybe a while.

• has intelligence truly been getting cheaper and cheaper? what’s the limit of large-small distillation? do we expect huge hardware efficiency gains that could further depress the price of intelligence? a: a lot.

• how much does latency matter? how much will people pay for low latency chips? a: a lot.

future geopolitics + race dynamics

• how focused is china? where are chinese labs getting money from, who are their backers, what is the government’s stance? what is the real risk from a geopolitical standoff vs misaligned exfiltrated ai vs social unrest from labor displacement?

• if rl capabilities and alignment are truly at odds, how do we coordinate against race dynamics? are race dynamics really a threat or maybe things are fine?

• is alignment easy if we allocate sufficient resources? yes, i hope so. given recent events, do we expect difficulty coordinating a slowdown? a: no.

• will the us government nationalize? what can/should we do about the social consensus around ai?

robotics, biology, and deployment

• what are my robotics timelines? what does robotics agi look like and what’s the evidence for when it will happen? what will be for robotics what hbm was for gpus?

• as isomorphic keeps getting better, and as anthropic doubles down on bio, what will the bottleneck in biology become? who’s solving it? what is the relationship to robotics?

• what is the future of quant? will we have ASI pod shops? if jane street can 100x alpha generation with automation, will they start squeezing into the small ponds? if we’re solving millenium problems why have we not solved hedge fund management? if market players gain accelerating capabilities does that not have a huge centralizing effect?

• where is energy today, and where is it going? a: i’m long panthalassa?

labor and social dynamics

• will companies layoff? which ones and when? is the citrini nightmare real?

• what will labor look like post agi? if people cannot define themselves by their occupation, what will they spend their time doing? what new things will become valuable to people at a personal level?

will keep this updated with more as they come to mind. unfortunately answering each of these requires a good amount of research so stay tuned.