zkOmelette

oblivious transfer / hyperspectral vision enjoyer / garbled circuits / sapere aude

[zkOmelette @zkomelette]

@MrEwanMorrison @exopodiatrist According to the crow, Building taller buildings spaced further apart would free up ground space with grass and trees and other soothing nature stuff like that, and give renters more space. The walk up tenements in euro cities at that time were dense dismal places to live.

[zkOmelette @zkomelette]

@0xmaddie_ dissipate energy, increase the entropy somewhere else

[zkOmelette @zkomelette]

@danrobinson good aesthetics

[zkOmelette @zkomelette]

@XorDev Node based editing is a massive cable management problem. I vastly prefer describing a data flow graph using just standard sequences of expressions, functions, flow control and assignment statements. With code autocomplete and instant syntax/type checking it's even better.

[zkOmelette @zkomelette]

@mattparlmer Intelligence emerges from the full form of the animal. Human intelligence is not general, it is very specific to our bodies and life cycle. Therefore, Snow Leopard is my answer - isolated, beautiful, and dangerous - attributes that pair well with intelligence.

[zkOmelette @zkomelette]

@sama Standardized tests just measure IQ which is largely an innate, heritable factor. It is no different than legacy admissions in that it gives implicit preference to children of a narrowly defined hereditary elite. Why not just lower the value of uni credentials instead?

[dynomight @dynomight7]

Rewrote this post which argues that Simpson's paradox is just the first layer of a deeper philosophical problem where we don't know how to define "bias" when each case is unique https://t.co/6ALdH8EyW2

[zkOmelette @zkomelette]

@topology_gg Amazing progress!

[zkOmelette @zkomelette]

If they ban training data from the web or art sites, it would be surprisingly easy to replace by just paying people to take pictures of interesting things.

[zkOmelette @zkomelette]

@Plinz blocking makes it impossible to see contradictory evidence (i.e. an annoying person posting something good), whereas unblocked annoying people continue to annoy on average. One never doubts decisions that are never contradicted by experience. plz don't block me

[zkOmelette @zkomelette]

@0xfbifemboy total agreement. you are the best in the bureau. we need more of this mindset in law enforcement.

[zkOmelette @zkomelette]

@dankrad @AnthonyLeeZhang Riemann–Stieltjes integral I've found quite useful (conceptually and occasionally practically) and I only learned about it from a course on functional analysis. Also measure theory is probability theory. Probability theory is very useful.

[zkOmelette @zkomelette]

So funny that much of the text in the bottom shelf aesthetics is comic sans..

[zkOmelette @zkomelette]

I really prefer the LAION-5B aesthetic score of around 2-3.
https://t.co/4ccTmqqp9Z The top-tier in "aesthetics" is already an annoying cliche.

[zkOmelette @zkomelette]

While single MD trajectories appear to be okay for jax/xla, I would expect that it will encounter difficulties similar to ray-trace when we wish to compute gradients of more challenging simulations like free-energy surfaces etc. Dr. jit has shown the way

[zkOmelette @zkomelette]

This also helps to understand why naive application of autograd to monte-carlo raytrace blows up - the compute graph for an average over millions of MC samples is a just
a bad-case for ML frameworks (at least I can't figure it out).

[zkOmelette @zkomelette]

The dr. jit paper on differentiable raytrace contains deep physical insights. The raytrace computes transport of radiance from lights to pixels, and the derivative transports gradient information from pixels to parameters. The deriv is just a different type of raytrace.

[zkOmelette @zkomelette]

Really enjoying jax-md - seeing very exciting possibilities for automatic differentiation of molecular dynamics and other similar simulations. https://t.co/XdGOSBy0gh

[zkOmelette @zkomelette]

The "gender" in the language of the AIs will be reed-solomon polynomials over large prime fields

[zkOmelette @zkomelette]

Gender in language is a mechanism for error detection / correction. Gender agreement rules are like a redundant check bit. No need to believe that the mind partitions all objects as m/f/n. The biggest problem in communication is the low apparent bit rate of natural language.