will grathwohl
@wgrathwohl
Followers
4K
Following
879
Statuses
2K
Lover of raccoons and machine learning. Daddy to Scumpo Drumpfus. Occasional cyclist for justice
New York, New York
Joined June 2010
Yilun cannot be stopped !!!
Excited to share our work combining video models and generative policies through MPC (Generative Predictive Control)! This combination allows us to get significantly more robust policy execution and downstream task completion.
0
0
13
I think I’m over machine learning. I’m more into AstroBot rn
1
0
6
@zdhnarsil @DrJimFan I think it’s the same plus the ppo clipping thing… I think this is a very good direction. If anyone understands pitfalls of training a neural network to be an effective control variate, it’s me! Def avoid if u can! I’ve been shocked many times how effective RLOO can be!!!
0
0
5
@drew_jaegle How’s the model output parameterized? Are you just predicting xt —> v ? @marikgoldstein prolly knows the answer to your question.
1
0
1
Y’all gotta get a dog. Desire to live x 10,000 since Scumpo Drumpfus was born.
0
0
20
This was made by my former team at GDM. They worked really hard on this for a long time. It’s so great to see it out there !!! Congrats
Google just released Gemma Embeddings! "GemmaEmbed is a dense-vector embedding model, trained especially for retrieval. As of December 12, 2024, GemmaEmbed achieves the #1 position overall on the MTEB leaderboard, with a score of 72.72."
1
0
11
Please god don’t tell me there is a vape detection system in Waymos
1
0
5
@sigmabayesian Yes. It will play you songs and talk to you and explain why songs are cool. I think it’s based on notebooklm
1
0
2
Mark rulez talk to the boy
(and I know this is a little early on the industry side, tell those postdoc fellowships not to have such early deadlines!)
0
0
3
Awesome stuff!! Hey @marikgoldstein how does this relate to the connections made in the stochastic interpolant literature??
A common question nowadays: Which is better, diffusion or flow matching? 🤔 Our answer: They’re two sides of the same coin. We wrote a blog post to show how diffusion models and Gaussian flow matching are equivalent. That’s great: It means you can use them interchangeably.
0
0
3
Hell yea
CAT3D + time => CAT4D! 🐈 Check out our latest work on turning text/image(s)/video into dynamic 3D models that one can explore in real time, led by brilliant @ChrisWu6080!
0
0
4