E. Z.

quantitative researcher... math and coffee are my life

[E. Z. @spectralquant]

@bryanrbeal @ThierryBorgeat Anyone that works for a fortune 500 knows management is retarded.

[E. Z. @spectralquant]

@jxnlco Optimization of CUDA kernels - Codex tries to do small changes while Claude will make big architectural changes that sometimes pay off bigly.

[E. Z. @spectralquant]

@whyarethis

[E. Z. @spectralquant]

@whyarethis Similar work I've been doing :)

[E. Z. @spectralquant]

@iamtexture It's such a simple reductio that at this point it just shows their delusion.

[E. Z. @spectralquant]

@iamtexture What is this... white noise sorcery? ;)

[E. Z. @spectralquant]

@allTheYud If thought was simply the action of atoms and electrons how would you ever know anything? It’s just random motion of electrons through your brain. It makes knowledge impossible. Therefore any such belief leads to a contradiction. QED.

[E. Z. @spectralquant]

Gotta chuckle at the LLMs sometimes

[E. Z. @spectralquant]

@rickasaurus On the other hand, there are people out there that think functions x_n+1 = f({x}_i=1^n) is alive. One of the groups is obviously delusional.

[E. Z. @spectralquant]

@JesusFerna7026 Yes. It's why FGLS is so important in economics yet not quite heard of elsewhere. One would hope that fixed point methods would be able to get to equilibrium faster, but I'm not sure that's easy to do in general.

[E. Z. @spectralquant]

@paleochristcon I just finished watching the entire whatever debate with NSE. Wow. You're a saint - assumptions require mind, but so does objective truth because we're assuming. I would've flipped the table at the retardation.

[E. Z. @spectralquant]

@teortaxesTex @lu_sichu @YouJiacheng It must be because of OLS

[E. Z. @spectralquant]

@hftgod Well... if the pot is literally going from P to 1/P then it's a quadratic I think with phi as solution? If the pot is pot - 1/pot then drift in the pot is ~0.5. depends on the reading.

[E. Z. @spectralquant]

@hftgod Ahh that changes everything

[E. Z. @spectralquant]

@hftgod Yeah - I think your nav is best seen as a random walk with drift 1/2 * (1/t - 1)

[E. Z. @spectralquant]

@hftgod Seems pretty clear you shouldn't pay: you win at most $1 from the house, the rest of the pot is your money. It's not a simple probability game it's a common sense one. You're bounded above by $1 nav. Kinda like a brownian motion with a take profit. You'll go negative pretty quick

[E. Z. @spectralquant]

@apathium65906 @0xfdf The problem is much more difficult than it seems.

[E. Z. @spectralquant]

@junhoBTC No - needs to have right elevations and proper pathfinding. No amount of water would make Colorado submerged for instance, because it would just go around it.

[E. Z. @spectralquant]

@systematicls If you can go from X->Y_hat->X_hat with minimal error, you have the optimal tracking subspace. Feature selection is solved.
Gauss was *so* good to us - he discovered and shared with us what can only be described as AGI. The problem is that this is most of us:

[E. Z. @spectralquant]

I think that sometimes it's ideal to go back to the basics and this is a case where I'd probably do that: use the actual distribution of the data and build the features as directed divergences straight from the distribution rather than go forward with that. Personally that's what I'd do...