Bart Wronski

What if The Matrix was only “accidentally” a masterpiece? (lucky shot) It was one of my formative movies (12yo when it released- perfect timing),but any subsequent work in the universe was ranging from confusing to terribly bad… other work of Wachowski sisters mediocre at best.

(respecting the original here, but Tiga and Zyntherius version is the one I heard first and subjectively better)

Is that because audio processing relies on discretized differential equations (analog filters etc) and discretization introduces significant error? Or maybe because cochlea directly measures frequencies through bandpass filters? Probably all of the above! Fascinatingly different.

Kind of interesting why? Is that due to nature of what we want to do with images vs audio signals? Less “interesting” stuff going on close to visual Nyquist? Way less sensitivity to visual aliasing (we are “used to”seeing rectangles and pixels)? We are Used to aliasing illusions?

Anytime I read about more complex audio processing, I am amazed how much easier *everything* gets with “oversampling”. But comparison in graphics / image processing, it’s much more limited - apart from antialiasing (duh), not many problems benefit from spatial supersampling.

Spectral rendering talks were one of my highlights of Siggraph, including this one. Highly recommend to appreciate how much it matters in practice - a lot. Three primaries and spectral aliasing (due to subsampling!) messes up colors badly, especially cross device boundaries.https://twitter.com/siggraph/status/1435668016142897156 …

Language differences, expectations: I remember when we just moved to US and saw “grilled cheese” everywhere on menus. We got excited - we love Halloumi and similar cheeses. You can imagine our disappointment when instead we got some FUCKING TOASTS.

Sparse, irregular, unordered data is everywhere - and in every domain. We need to go beyond brute-force regular discretions (and I don’t think that continuous coordinate based are a solution for most problems). Useful to have some tools and mental frameworks for those.

I highly recommend the post even if you don’t care about NNs - makes you think . But also beautifully and visually explained, with examples and color coded equations. :)

Another problem is connection count invariance. Images are simple- every pixel has 8 neighbors. What if someone has 4 vs 40 social connections? Or mesh vertex with 2 vs 90 neighbors? How do you design a single algorithm handling those? Graph Laplacians and spectral decomposition.

Images are graphs. Meshes are graphs. Social connections are graphs. Sports team rankings are graphs. “Reality” is usually not ordered. If you compare team A to team B, it shouldn’t care about which one you entered first.

It's also much easier to get into making music. You can get a *fantastic* guitar sub $100,free internet content teaching you everything. Or if you prefer producing, your computer or even phone is enough to make Grammy winning tracks. Just matter of ideas and work you put into it.

This is not to take away from nostalgia and my love for older music (whether 60s or early 00s). But I really don't get "boomer talk". So many new genres, experimental stuff, artists, everything so accessible. Complaining they don't play x/y/z on the radio is intellectual laziness