cool. would be nice to see some benchmarks~

We’ve developed Gemini Diffusion: our state-of-the-art text diffusion model.

Instead of predicting text directly, it learns to generate outputs by refining noise, step-by-step. This helps it excel at coding and math, where it can iterate over solutions quickly. #GoogleIO https://t.co/EmDAqNeXVv

Google DeepMind

cool. would be nice to see some benchmarks~

https://x.com/googledeepmind/status/1924888095448825893?s=46&t=bXRj3LXXGHjoj1idpl6gdQ

vibe coding @farcaster || prev BS/MS Berkeley EECS

interesting~ dont know too much about them. looks like it's focus for coding

the thing is im generally a bit bearish on diffusion for text generation. No doubt it works well for images/videos, but diffusion model for text 1) more expensive 2) benchmark is marginal/meh relative to transformers

i wonder how this compares to Inception’s models https://www.inceptionlabs.ai/introducing-mercury

I’m bullish on diffusion models for text, if they work the speedup is going to be huge