Twitxy Saved

Orr Zohar

Verified account This account is verified. Learn more

@orr_zohar

@nvidia • @Stanford • @KnightHennessy scholar • Researching AI that understands the physical world
245 Following 646 Followers 164 Posts

Latest posts 5

Pinned post

Orr Zohar Verified account @orr_zohar
FineVision is out 🚀

And ready to empower the next generation of LMMs.

Check it out!
Luis Verified account @lusxvr Sep 4, 2025
Today, we are releasing FineVision, a huge open-source dataset for training state-of-the-art Vision-Language Models:

> 17.3M images
> 24.3M samples
> 88.9M turns
> 9.5B answer tokens

Here are my favourite findings:
View quoted post on X
1 replies 4 reposts 21 likes 6.1K views View on X
Orr Zohar Verified account @orr_zohar
Working with @tokenpilled65B is always fun :) over a year ago i remember him complaining about the perceptual losses.

Indeed, it is time for an update!
Philippe Hansen-Estruch Verified account @tokenpilled65B Dec 30, 2025
LPIPS has been the go-to perceptual loss for autoencoder training, but it's built on VGG features from 2014.

Today we're releasing DINO Perceptual Loss: a modern replacement using DINO that achieves better results without adversarial training:

uv pip install dino-perceptual
View quoted post on X
Orr Zohar Verified account @orr_zohar
People underestimate how token-heavy video understanding & generation are - and their massive untapped potential.
We grasp images in a split second. Videos? Not so much: Comprehending a 1-hour clip often demands watching the whole thing.
Video AI is a tiny sliver of tokens today, but I predict it'll explode over the next 1-2 years as models level up, potentially eclipsing text-based apps in usage. 🚀
Elon Musk Verified account @elonmusk Nov 7, 2025
Diffusion will obviously work on any bitstream.

With text, since humans read from first word to last, there is just the question of whether the delay to first sentence for diffusion is worth it.

That said, the vast majority of AI workload will be video understanding and generation, so good chance diffusion is the biggest winner overall.

Also means that the ratio of compute to memory bandwidth will increase.
View quoted post on X