Much has been said about many companies’ desire for more compute (as well as data) to train larger foundation models. I think it’s under-appreciated that we have nowhere near enough compute available for inference on foundation models as well. Years ago, when I was leading teams
AI HARDWARE
-
NVIDIA Aerial Omniverse enables 5G 6G R&D digital twin simulation
By
–
Learn how R&D on #5G and #6G and optimized network planning is enabled through NVIDIA Aerial™ Omniverse™ Digital Twin—a next-generation, system-level simulation platform now available in the NVIDIA 6G Research Cloud platform. Explore the platform. https://
nvda.ws/3vYcEc1 -
Apple’s Spatial Computing Lacks Imagination Despite 3D Portal Potential
By
–
The most disappointing thing is ultimately that Apple made every single experience so boring. You can create 3D portals to anywhere and they decided the only thing people want to do is to create multiple virtual 2D computer monitors. Amazing possibilities, but no imagination.
-
Apple Vision Pro Lacks Killer Apps for VR Launch Success
By
–
It is weird that, after all the investment, Apple hasn’t spent the money for more super high quality apps to support the launch. There are no social experiences at all (the killer VR feature), no games, no communal workspaces. Just the world’s most awkward Zoom implementation.
-
GradCache Optimization for Distributed GPU Training
By
–
oh yeah. I think his trick was just GradCache. and sharing negatives between GPUs isn’t trivial
-
Screen-less devices unsuitable for shopping use cases
By
–
The reason I don't order stuff on Alexa is exactly the same as why I wouldn't order it on the Rabbit R1. Shopping is a terrible use case for products with no screen imo.
-
NVIDIA Delivers First DGX H200 to OpenAI for AI Advancement
By
–
First @NVIDIA DGX H200 in the world, hand-delivered to OpenAI and dedicated by Jensen "to advance AI, computing, and humanity":
-
Enterprise AI at Scale: GPU Infrastructure Challenges
By
–
Transforming Enterprise AI at Scale "Scale sets a speed limit; if you don't have scale, you can't train some of these massive models." In a recent interview with Sequoia, Andrej Karpathy mentions this and highlights the difficulty with instrumenting large GPU clusters that
-
Local LLMs on Phones Connecting to Cloud Intelligence
By
–
Pretty clear signaling from the flurry of papers by Microsoft and Apple that local LLMs on phones connecting to bigger LLMs in the cloud are coming. Phi 3 is already far better than Siri or Alexa, and with tools and the ability to call on a smarter intelligence, it can do a lot
-
Apple Compatibility Challenges for AI Implementation
By
–
I haven't managed to run it on an Apple machine yet!