randomgermanguy

But why do you think this is? Like is it just the money/status that comes with calling yourself an "AI"-expert ?

If the alternative is a Linux-distro, likely UX won't be much better/more-consistent when applications use different UI kits/styles etc.

Even Though Apple is doing a shitty job with their walled garden, a garden is still more organized than a jungle of different distro's/applications/frameworks/etc.

(at least in my limited experience)

Made me instantly think of this: https://www.nationalgeographic.com/science/article/slime-mou...

Something ironically self-referential about not only fungi growing (like) rail networks, but also IN rail networks.

Depends on how heavy one wants to go with the quants (for Q6-Q4 the AMD Ryzen AI MAX chips seem better/cheaper way to get started).

Also the Mac Studio is a bit hampered by its low compute-power, meaning you really can't use a 100b+ dense model, only MoE feasibly without getting multi minute prompt-processing times (assuming 500+ tokens etc.)

I think the only exception is specifically for studying network/communciation-topologies. I've seen a couple clusters (ca. 10-50 Pi's) in universities for both research and teaching.

I personally just throw them at the top of my files as long block-comments, or sometimes inside/around very heavy functions. For example i often add little diagrams for when dealing with some bit-fiddly logic parts to easier visualize the bit-layouts. But for architecture, either a whole text-file for it or at the top of the module

As the other commenters put it, i dont think this is a huge issue. I usually use this for architecture level diagrams, and that shouldn't change often/at-all. In-case it does change, doing a new diagram is perfectly in-scope of whoevers working on that.

Bought this couple months ago, and am now always looking for more ways to include this for inline-documentation.

the fact i can export to clipboard and re-import it and reconstruct all the shapes etc. almost flawlessly is such a big win.

Funnily, they're far from being optimal for GEMM ops (especially in terms of power consumption).

For GEMM you need to visit each row/vec n-times so theres a bunch of data-reuse going on, which isn't optimal for GPUs since you can't keep that all so close to your processing-units. And while the tensor-cores kinda implement this i think they don't quite scale up to a full sized systolic array, which is you would want for larger matrix multiplications.

Also just a simpler view: with GPUs most of their silicon is spent NOT tensor-core, so just from that you know its not optimal i guess.

Just referring to that FLOP/s number doesn't really mean much nowadays with tensor-cores and sparsity.

In my eyes the big win of GPUs are that not only are they pretty good at GEMMs but also really good at a lot of other easily parallelizable tasks PLUS they're comparatively easy to program ^^

Spherical harmonics are used all over the place in climate science. But two big examples are Global oscillations like NAO&ENSO (so ocean currents and atmospheric pressure mostly), and analyzing wind-patterns on the globe/sphere.