my notes on whatever i find interesting β¨ π¬ science-backed π curiosity driven π€ host on @SciShow@KQED@PBSTerra@FeministBookClub π san francisco
Your brain deserves better than brain rot π§
Come learn with Big Ideas: compound eyes π eggs π₯ bioluminescence β¨
Human made. Real scientists. Real filming.
Out now: youtube.com/watch?v=5UCQoFO9β¦
from the @kqedscience team behind @deeplook
This indie dev is making a chess game where every piece has been replaced by a cat with its own personality and breed
- Stray cats vs house cats
- Play with your cats while waiting for the opponentβs turn
- The purrfect chess game
Would you play this? It's called Cat Chess.
1/2 "It seems silly, but it's a very important problem." Virginia Vassilevska Williams (@MIT) on the progress in matrix multiplication algorithms during her Richard M. Karp Distinguished Lecture On Matrix Multiplication Algorithms at the Simons Institute. simons.berkeley.edu/talks/viβ¦
π Started my Science Communicator Residency at @UCBerkeleyβs @SimonsInstitute ! I'm learning how advances in matrix multiplication shape the future of AI and turning it into a mini documentary.
π₯ A peek into my first day:
Nearly everything in modern computing depends on matrix multiplication and researchers are still chasing the fastest way to do it.
Virginia Vassilevska Williams helped break this speed record, then turned to understanding why some barriers canβt be broken at all.
@SimonsInstitute
Prof Williams now studies when speed runs out.
Her field, fine-grained complexity, reveals how close we are to the limits of computation and why some problems resist every shortcut.
@SimonsInstitute#SimonsLive
This week Iβm at a matrix multiplication workshop (me, not a mathematician or computer programmer) because multiplying matrices of numbers sits behind every single AI model and every big data crunch, and I want to understand these systems better. @SimonsInstitute#SimonsLive
Avoiding unnecessary data movement saves energy too!
Holtz showed how communication-avoiding algorithms can find a sweet spot where less motion β> lower energy use without slowing computation.
@SimonsInstitute#SimonsLive
Most of a computerβs effort isnβt doing math, itβs moving data, aka every trip between memory & processor = costs in time & power
This is the communication cost, a framework researchers use to measure and reduce that hidden expense.
from Olga Holtz @SimonsInstitute#SimonsLive
How to make matrix multiplication practical
Mathematically:
- shrink the exponents and constants that set its speed
Hardware:
- move data less
- handle smaller chunks more efficiently
(from Oded Schwartz) @SimonsInstitute#SimonsLive
If we can multiply matrices even a little faster, the effects ripple outward. AI models could train more quickly, supercomputers could run with less energy, and data centers might draw less power from the grid.(Efficiency here is environmental as much as intellectual) #SimonsLive
Matrix multiplication is the engine of modern computation. It powers neural networks, graphics, and massive data analysis. Every improvement changes what computers can do within the same energy and hardware limits. #SimonsLive
Iβll be here for the next month, following how researchers rethink one of the most fundamental operations in computing and how a small tweak in efficiency could reshape technology. #SimonsLive
with all the social media apps slowly morphing into each other, youβd think thereβd be more tools to just cross post things everywhere but ohhhh nooooo the kids canβt play together
from filming SciShow to habit tracking to my banana bday party πΒ tryna chase perfect productivity but thereβs always more to do, so i just gotta ride a wave of momentum π π¬ π§³Β #weekinmylifeyoutu.be/8061mzwrYAQ