We’re excited to be heading to CVPR 2026 next week.
Over the past year, Sieve has built one of the largest repositories of multimodal data and evals frontier labs use to train and evaluate world models, video generation systems, and robot policies.
If you are a researcher or builder working in the multimodal space, drop us a message here or at [email protected] — we'd love to connect at CVPR!
we hosted a multimodal researcher lunch at @ycombinator with the @sievedata team, who've been doing really good work in multimodal data
we got researchers from several frontier labs in a room together and the conversation ranged from robotics to world models to computer use agents. More of these coming soon!
@ericjang11 congrats Eric, it's been awesome to listen to all ideas you're exploring these last few weeks. I'm sure whatever comes next will be incredible :)
Two beginnings of emergent reasoning from scaling:
- 2022: "let's think step by step"
- 2025: "generate a video of"
Emergent reasoning is a necessary and sufficient condition for AGI and ASI. We shall figure out some nice scaling for continual learning too.
Our third announcement is GWM Worlds. A world model for real-time environment simulation.
You give the model a static scene, and it generates an immersive, infinite, explorable space as you move through it, with geometry, lighting and physics. All in real-time. You can travel to any place, real or imagined. You can become any agent, a person walking through a city, a drone flying over a snowy mountain or a robot navigating a warehouse.
Say hi to Odyssey-2 Pro—our most intelligent, interactive, and general-purpose world model yet.
Every pixel is generated in real-time, and frame-by-frame the model simulates the world forward. Today for minutes, soon for hours.
Available to developers and our API very soon.
While I have great fondness for @NASA, they will constitute less than 5% of our revenue next year. Commercial Starlink is by far our largest contributor to revenue.
Some people have claimed that SpaceX gets “subsidized” by NASA. This is absolutely false.
The SpaceX team won the NASA contracts because we offered the best product at the lowest price. BOTH best product AND lowest cost. With regard to astronaut transport, SpaceX is currently the only option that passes NASA safety standards.
I'll be a NeurIPS 3rd-5th this coming week. DM if you want to chat video generation, world models, and the interesting ways in which to collect and curate data for them.
Interesting to think about how crypto mining kicked off limitless demand for Nvidia chips, datacenters+electricity and how neatly it translated to AI.
If Nvidia had no crypto mining business what state would they have been in to respond to demand from models?
New w/ @nmasc_@MilesKruppa: Richard Socher, CEO of search startup https://t.co/6qDswrQ1uj, is raising $1B for a new AI lab developing tech to automate AI research.
He'll remain CEO at https://t.co/6qDswrQ1uj, he tells us.
https://t.co/WNbLijm8dA
.@ericzelikman & 7th Googler @gharik are raising $1b for an AI lab called Humans&. I'm told Eric's paper STaR was an inspiration for OpenAI's reasoning models, and that he was also one of the star AI researchers labs fought over. https://t.co/nrgEpcL6Mc
Raymond James just said the AI cloud will be a $350B market by 2030.
They think the split looks like this:
• $MSFT 37%
• $ORCL 35%
• $CRWV 11%
• $GOOGL 8%
• $AMZN 7%
• $NBIS 3%
The wild part is Amazon at only 7%.
AWS still has the largest base of enterprise relationships, deepest cloud integration & biggest commerce engine behind it. If even a small slice of that base shifts AI workloads onto Bedrock + Trainium the rerate will be obvious.
I'm buying the gap between what people think Amazon is and what it is actually building and that is why Amazon is now my biggest position.