1/7 Wondered what happens when you permute the layers of a language model? In our recent paper with Max Tegmark, we swap and delete entire layers to understand how models perform inference - in doing so we see signs of four universal stages of inference!

Stefano Ermon Inception Diffusion will obviously work on any bitstream. With text, since humans read from first word to last, there is just the question of whether the delay to first sentence for diffusion is worth it. That said, the vast majority of AI workload will be video understanding and

Excited to have this work included in the PEFT library! PR: github.com/huggingface/pe…

"It's now or never." Our CEO @stefanoermon on why he started Inception – featured this week in The Wall Street Journal. We're building dLLMs that generate tokens in parallel. Faster. More efficient. More controllable. This is the moment. Thanks for the coverage Kate Clark

Diffusion clicked for me when I read about score-based models, a line of work pioneered by Stefano Ermon (et al.) at Stanford. So it was a full-circle moment to collab with him and Inception on a video about training & sampling techniques for making diffusion LLMs faster.

The more structure a language has, the faster diffusion can run. Code fits that profile. Code has plenty of it. Listen to Sid Kharbanda on how diffusion unlocks speed for real-world coding workloads. #Diffusion #AIInfrastructure #DeveloperTools