GLM 5.2 is an MoE, NVFP4 is 467 GB, and the DGX Station comes with 496GB LPDDR5X + 252GB HBM3e GPU memory With the right offloading formula, it should work
MACHINE LEARNING
-
Working on getting GLM 5.2 NVFP4 by Luke Alonso running
By
–
Currently working on getting GLM 5.2 NVFP4 by Luke Alonso up and running 🙂 Will report back
-
NVIDIA-accelerated AI aids PYLER in brand safety for advertisers
By
–
Every day, millions of videos compete for advertising dollars. Ensuring brands appear alongside the right content requires AI that can understand context at scale.
— NVIDIA (@nvidia) 25 juin 2026
PYLER is helping advertisers improve brand safety and campaign performance with NVIDIA-accelerated AI that analyzes… pic.twitter.com/9xSDjj9e9gEvery day, millions of videos compete for advertising dollars. Ensuring brands appear alongside the right content requires AI that can understand context at scale. PYLER is helping advertisers improve brand safety and campaign performance with NVIDIA-accelerated AI that analyzes
-
Pangram false positives; skepticism about trusting numbers
By
–
FYI Pangram is a coin toss, i've got crazy false positives with it
And this text does not smell AI-generated anyway (believing something just because there's a number on it is a bit midwitical tbh) -
ViT³: Test-Time Training Replaces Attention with Online Learning
By
–
Why settle for attention when you can learn at test time? Tsinghua University & Alibaba Group present ViT³: a pure Test-Time Training (TTT) architecture that replaces attention with an online learning model built from key-value pairs. This inner model trains on the fly,
-
Pim de Witte accidentally built the perfect world model data collection business
By
–
on their @latentspacepod we covered how @pimdewitte accidentally made the PERFECT world model data collection business by collecting the world's largest dataset of trainable (video,action) pairs.
— swyx @aiDotEngineer WF (@swyx) 25 juin 2026
turning the attention economy into the attention industry.
congrats Pim!!
link… https://t.co/Al4NGdX71W pic.twitter.com/opgI2Kk95Kon their @latentspacepod we covered how @pimdewitte accidentally made the PERFECT world model data collection business by collecting the world's largest dataset of trainable (video,action) pairs. turning the attention economy into the attention industry. congrats Pim!! link
-
No solution to AI hallucinations despite years of promises
By
–
Crazy how many times people have told me over the last five years that a solution to hallucinations was right around the corner — and yet here we still are.
-
Why Big Labs Ignore Continual Learning: It Runs Locally
By
–
Continual Learning will run locally That's why the big labs aren't talking about it Not your weights, not your model, LITERALLY
-
Midjourney updates: –preview for V8.2 and –sref random batch mode
By
–
Two quick updates in image world. Try adding –preview to your prompt for a early peak at V8.2 aesthetics & personalization. We've also updated our big batch draft mode to work with –sref random so you can explore style space 24x faster than before. Enjoy!
