interesting writeup. funnily enough, in universal paperclip maximizer's midgame, thats exactly what you do. you do ai research and your flywheel is using ai to make money in the hedge fund you run, which helps train better ai
i've always felt that lab flywheels are so much more complicated compared to a hedge fund flywheel. labs require consumer/business integrations, product building, marketng, fundraising etc whereas a hedge fund, theoretically requires just market access and it can instantly tap into the largest pool of money and bootstrap from scratch
i ... (read more)
Did you read the recent Self Hosting post? You might resonate with it
I agree with the possibility you've described although, I disagree with the key takeaway from it.
Your post takes away: There is an RSI-enabled AGI possible that progresses gradually before reaching a point of rapid takeoff.
What I think is more important: There is a strict lower limit on how fast we take off now, because even the current loop of RLVR with non-AGI strength models is already AGI. That takeoff is inevitable.
agreed, the fundamental "self-recursive" model of twitter feels more complete. twitter's UI of quickly understanding context is abysmal. fun fact, everything on Reddit is also a single data model "Post". posts/comments are just rendered differently
agreed, but these problems are UI problems. not a critique of the concept of being "fully recursive" that LW is not
We use chore markets! We have a board on the fridge where we log chores as we do them, then settle up every ~10 days, trading off against each other. Since we perceive the burden of chores differently, the "prices" aren't fixed: we negotiate and adjust them over time. e.g., unloading the dishwasher started at ~1.33x taking out the trash; it's now ~2x, because the dishwasher needs to be done more frequently (more repetitive = more chore-like) and with more timeliness (more urgent = more chore-like)
Some advantages I've observed
You're assuming that there will be a distinction post-singularity? Are you not a physicalist?
agreed, pretty much the same thing as thinking in neuralese vs thinking in human language