The purpose of that is to reduce memory bandwidth requirements, I assume?
@jeremyphoward
-
FFT and Winograd algorithms for neural network acceleration
By
–
Are you guys using fft, winograd, etc nowadays to speed it up?
-
E-fuse Limitation Impact on BF16 to FP32 Accumulation Speed
By
–
I think this is the e-fuse issue isn't it, which is artificially limiting the bf16->fp32 accum speed? So strictly speaking "same chip" is correct, if a bit misleading?…
-
Young Programmer Learning p5js, Scratch, and Python Skills
By
–
My daughter does a *lot* of p5js (many hours every week), and still kinda likes scratch. She's been doing a bit of python in Solveit. She hasn't been into the robotics stuff for a few years now — guess she takes after me in liking software. 🙂
-
Learning World History Culture Politics Intrinsic Value Knowledge
By
–
It includes history, along with culture, and politics. Seems fine to be interested in learning stuff about the world? Not everything should be expressed in terms of an opportunity cost. Sometimes advancing one's knowledge has intrinsic value anyway.
-
OpenAI Customer Questions Product Progress Despite Advanced Models
By
–
But I'm referring to the products they're making with the magic box. Shouldn't they be getting much better much faster than before? I don't feel like I'm seeing great progress, as an enthusiastic OpenAI customer.
-
OpenAI’s Product Development Claims Questioned by Skeptics
By
–
You'd think if OpenAI has transformed how they build in a way that actually ships better products faster, customers would be the ones to notice, and we wouldn't need to be told…?
-
Hetzner Infrastructure: Understanding Foundational Technology Matters
By
–
Definitely Hetzner. No way I'd want a team that didn't understand and couldn't work directly with actual foundational tech.
-
Douglas Adams and Reinforcement Learning for Robotics
By
–
Douglas Adams invented reinforcement learning for robotics in 1992 🙂 https://
en.wikipedia.org/wiki/Mostly_Ha
rmless
…