I thought it was a neat (even if not super practical) observation: we often think of clean, closed form expressions as the best way to represent the solution of a problem ("it's obviously O(1)!") but that's not always true for every context

It turns out the bottleneck is the fp multiplication itself: it operates on BigNum-style floats and requires computing an FFT (or two?) per multiply. And you have to do a lot of these to exponentiate, so the cost adds up quickly

This came up in a video I watched recently! It was about computing Fib() when N is extremely large - in the thousands of digits or more - and how to do it efficiently. The exact formula was one of a few methods they tried and it didn't perform that well, surprisingly

i've heard so many good things about rhythm doctor, excited!!

its not a direct answer to your question, but in the case that your array represents the cdf of a tabulated function and you want to invert it in order to sample from it, then there is a beautiful o(1) method for *that* called vose/walker's alias method, iirc. not quite your q, but beautiful

yep, this is the same solution I wound up on to do e.g. inverse CDF sampling ^^

GMM!!! i can’t believe she still recognizes that comforter - i guess some things never change 😌

(it looks super cozy fwiw)

The video:

youtu.be/il-TXbn5iMA

I also really love this excerpt from the opening of his book on Figure Drawing. Loomis wants you to know that you matter as an individual -because- art is about thought, intention, humanity. You can steal an artists' style but not their soul or their heart.

i know right!! iirc it’s the one where they’re gonna score it alongside a live speedrun attempt? such a neat idea for a concert and i’m so bummed i can’t go

🫂 take care