_wsa's profile picture. https://wsa.ai/

wsa

@_wsa

https://wsa.ai/

How, indeed. (This, we can agree on.)

I don't understand the math culture wars. The right says they want everybody regardless of background to meet the same high performance standards. The left says they want to raise teaching quality for traditionally underperforming groups How are these not the same position???



R + tidyverse has, hands down, the best syntax for working w/ structured & tabular data so far, and it's as fast as R can be. There's just no comparison. The kernel of truth here is: limiting dependencies means often having to live outside the tidyverse. But c'mon . . .

I don't understand the math culture wars. The right says they want everybody regardless of background to meet the same high performance standards. The left says they want to raise teaching quality for traditionally underperforming groups How are these not the same position???



wsa reposted

When it comes to similarities between the brain and deep learning, what's really striking is that everything that was actually bio inspired (e.g. sigmoid/tanh activations, spiking NNs, hebbian learning, etc.) had been dropped, while... (Cont.)


Yikes.

2022 ended with a shiver, but the start of 2023 broke a sweat. 🥶🥵 This weather whiplash may have been caused by changes in the polar vortex, the band of strong westerly winds over the Arctic. Details: go.nasa.gov/3XaD4iK



Intuitionist / positivist?

wait how did i miss this anti-AI and pro-AI artists are ____ and ____-coded, respectively



Facts.

wait how did i miss this anti-AI and pro-AI artists are ____ and ____-coded, respectively



Wait till you hear about nonstandard models of Peano arithmetic . . . turns out, in a reasonable theory of the naturals, you can’t just add a couple of things and be done there like you can with the reals.

Learning about nonstandard analysis and it’s just…beautiful. Somebody decided analysis wasn’t cursed enough and was like “you know what would make this better? That’s right, model theory.”



wsa reposted

How good are today’s foundation models? Stanford CRFM scholars evaluated 30 language models from 12 organizations as part of their new HELM benchmark. They found: 🧵

StanfordHAI's tweet image. How good are today’s foundation models? Stanford CRFM scholars evaluated 30 language models from 12 organizations as part of their new HELM benchmark. They found:
🧵

wsa reposted

❓ Did Diffusion meet Segmentation? 💡 The answer is YES! ✨ MedSegDiff: Medical Image Segmentation with Diffusion Probabilistic Model It performs even better than the existing models when tested with medical images

ai_fast_track's tweet image. ❓ Did Diffusion meet Segmentation? 

💡 The answer is YES!

✨ MedSegDiff: Medical Image Segmentation with Diffusion Probabilistic Model

It performs even better than the existing models when tested with medical images

Most of the code I write is meant to be run a small number of times to create persistent artifacts that can used by ‘real’ systems, Python seems now to be the default choice of the ML community, and it is just fine for my purposes 99.9% of the time. But it . . . has issues:

1/ Many will tell you why Python is great for teaching coding, so I'll tell you ways it's not. State is a bad default. It should be legal but safe & rare. The arc of programming is long and bends towards immutability. Its early use creates messes (eg, "a variable is a box".) ↵



This account does not follow anyone
Loading...

Something went wrong.


Something went wrong.