Unesd

🇲🇦 Dedicated to A.I and Tomography.

[Unesd @YNess97]

@Achl7i555 chfto b7ala skipa l3obodial fl7adara l2islamia

[Unesd @YNess97]

@miniapeur The prompt?

[Unesd @YNess97]

@prof_kamilov Funny how “professor @ major US university” disappears from the bad stories 😁

[Unesd @YNess97]

@prof_kamilov Let me reassure you that you do have a fan base! My officemate and I are big fans of your work, and we regularly stalk your LinkedIn.. waiting for your next preprint!😁

[Unesd @YNess97]

@prof_kamilov And for postdocs?

[Unesd @YNess97]

@Achl7i555 الشوط*

[Unesd @YNess97]

@Achl7i555 الحاجة الوحيدة لي مادرناش فهاذ الماتش هي ماهزيناش عل لاقل ١

[Unesd @YNess97]

@LiangZheng_06 Fair, but isn’t the hardness partly just shifted to the autoencoder training?

[Unesd @YNess97]

@_suthar_mayank_ @ranaharshraj7 @NVIDIAAI Thanks.. it looks awesome

[Unesd @YNess97]

@GordonWetzstein 4/4
Could you clarify where u lives in PIXEL-DC?

Is u intended to be latent-space, with the shift applied before decoding, e.g. D(z̃−u)?

Or is u pixel-space in the PIXEL-DC branch, with an implicit mapping back to latent space?

[Unesd @YNess97]

@GordonWetzstein 3/4
But later, u appears in the latent reverse update:

z_{t−1} ← ... + sqrt(ᾱ_{t−1})u

and in the dual update:

u ← u + x − z̃,

where x = E(x̂), so x, z̃, and z_{t−1} are latent-space variables.

[Unesd @YNess97]

@karpathy we will wait for nano-claude / nano-mythos video😁

[Unesd @YNess97]

@e_pavarotti 396?

[Unesd @YNess97]

@TheAhmadOsman no one will understand this pov.. till trying a cluster..
u can launch multiple jobs in parallel and do a lot of experiment at the same time!
the only drawback is the $$ to build it!

[Unesd @YNess97]

@benomar555 داكشي لي دار مع الوداد؟ و تا الفتح؟

[Unesd @YNess97]

@TheAhmadOsman Fast, and u can train on them from scratch!

[Unesd @YNess97]

@VraserX @sama It is already omni?

[Unesd @YNess97]

@NielsRogge pure waste of gpu's and electricity!