4h ago

Inception highlights debate from this week's MLSys conference on whether large language models will rely on autoregressive methods or shift to diffusion-based approaches over the next decade

Kuleshov notes diffusion could replace sequential inference after parallel pre-training

——0——
Original post
Stefano Ermon#129@STEFANOERMONOPInceptionInception@_INCEPTION_AI

Will the next decade of LLMs run on autoregression, or on diffusion? One of the top questions we got at MLSys this week. Part 6, the final part of our founder story series with @timt at @MenloVentures. Featuring @StefanoErmon, @adityagrover_, @volokuleshov

10:32 AM · May 22, 2026 View on X
QUOTE POSTVolodymyr Kuleshov 🇺🇦#539Volodymyr Kuleshov 🇺🇦@VOLOKULESHOV

The transformer made training parallel and unlocked scaling laws for pre-training. Inference is still sequential. Diffusion is the evolution for inference.

InceptionInception@_inception_ai

Will the next decade of LLMs run on autoregression, or on diffusion? One of the top questions we got at MLSys this week. Part 6, the final part of our founder story series with @timt at @MenloVentures. Featuring @StefanoErmon, @adityagrover_, @volokuleshov

5:32 PM · May 22, 2026 · 2.6K Views
6:58 PM · May 22, 2026 · 847 Views