43%!

Many bad arguments are made by taking instantaneous non ergodic measures of dynamical systems. Throwing out the manifold on which a function is defined to talk about the fn is so dumb. If a system has direction and magnitude, and its behavior temporally posed- don’t remove them from discussion.

dude Pythagoras would be so ready and on-board to worship some vectors, his machine learning papers would be so amazing

Forget leetcode - the most dangerous ML interview question is what area of applied math did you pick up on the job. Everyone who isn’t picking up information theory/algebraic topology/numerical analysis/etc isn’t abusing the fringe benefits of the career/is lying and is should be seen as suspect.

I don’t have time to sit down and write it down but I’m almost certain pi-calculus and some ergodic measures over the described processes could partition signal source side human vs. algorithm side computations better than whatever is done here alone atm. Bots look parameterized to heck.

…have we collectively AGI hype sobered up enough that self supervised and energy based model research is getting published again?

I haven’t looked in a year or two but all that fell off a cliff in 2022 and it made me sad. #machinelearning

If I wanted to construct a financial instrument to find and then short stranded coal plants - where would I find the relevant data? Where are the amortization schedules, etc?