Nice, short post illustrating how simple text (discrete) diffusion can be. Diffusion (i.e. parallel, iterated denoising, top) is the pervasive generative paradigm in image/video, but autoregression (i.e. go left to right bottom) is the dominant paradigm in text. For audio I've

Aged like fine wine

Me: open laptop See this commit from Peter Tong Close laptop I know he cooked overnight and I just need to wait.

Great by talk by Luke Zettlemoyer summarizing our team’s work. We have more on the way this year ⏲️

MLLMs are great at understanding videos, but struggle with spatial reasoning—like estimating distances or tracking objects across time. the bottleneck? getting precise 3D spatial annotations on real videos is expensive and error-prone. introducing SIMS-V 🤖 [1/n]

Pre 2021: submit to a conference and on arxiv 2021-2024: submit to arxiv, make a website and tweet about 2025: all of the above plus promo video

If a benchmark can be gamed, it will be. Looking at you GenEval, DPGBench

Human reasoning requires the ability to revise and self-correct through multiple rounds of edits neither do AR or (masked) Diffusion LMs do this.

Today we are launching Marble – a multimodal world model that lets you create and edit 3D worlds.