Policy, pixel world model, latent world model, or world action model: what should a robot model predict? Mon Oct 26 at #MAINeducational in Montréal: a hands-on tutorial where we build all four on the same robot data. https://www.eventbrite.ca/e/main-educational-2026-tickets-1999327932667

Show Post

Switching between AdamW and Muon optimizers in LLM training incurs minimal costs. Final performance surprisingly improves with more Muon updates in fine-tuning, challenging the view that pretraining and post-training optimizers must match. https://arxiv.org/abs/2610.22650

Show Post

How to Train Your Robot | Discussion

Show Post

👇

Show Post

Anyone figured out how to enter a flow state while using agents? Still seems like an open challenge.

Show Post