Fuck Elon Musk, find me on bsky. https://xitter.gq/t.co/8V9JMr7Av3
Menlo Park, CA
Joined April 2008
- Tweets22.1K
- Following1.4K
- Followers2.1K
- Likes19.1K
iandanforth 🦋 @iandanforth.bsky.social retweeted
Bravo
youtu.be/UytYrhYLOKI
iandanforth 🦋 @iandanforth.bsky.social retweeted
In 'Living On Earth' I quickly survey tool use by marine animals and note how limited it is. Dolphin use of sponges to protect the nose is one case. Via @BBC, their use of sponges turns out to be richer. Very good video from the new Blue Planet show.
youtube.com/watch?v=q0QFv4iM…
iandanforth 🦋 @iandanforth.bsky.social retweeted
Simulation, world models, lab robots, robot gyms etc are all just proxies for deployment in the end.
In order to use any of them well you need to deploy asap. Otherwise you have lots of data - more expensive in some cases than others but never free (including the compute to train).
LLMs are great because deployment is easy, even coding is. You learn about the value of your proxies quickly, and can analyze how to tune them better going forward.
But just because robotics is hard does not mean you can arbitrarily delay deployment. The most important things about robotics startups right now is not the pitch of the base tech, but the pitch about how to apply it so that you actually test what matters - monkey first!
🤖 Made with AI
iandanforth 🦋 @iandanforth.bsky.social retweeted
We made the first AI model you can have a realistic conversation with that was trained entirely on speech!
Dessa is a full-duplex speech-to-speech model that you can run on consumer GPUs.
This is a GPT3 moment for speech.
iandanforth 🦋 @iandanforth.bsky.social retweeted
We found a way to scale humanoid control that reduces GPU-hours by more than 1000×.
Across 129,514 motions, SuperMimic surpasses the state of the art (128 GPUs, 7 days) with 1 GPU in 3 hours.
Project: supermimic.github.io
Demo: supermimic.github.io/demo/
iandanforth 🦋 @iandanforth.bsky.social retweeted
At an election science conference I went to in June, a panel was lit up by claims that USPS delays in primaries were causing more ballots to arrive late. Someone asked: Has anyone done research yet to see if USPS changes are tied to ballot rejections?
So @propublica did. 🧵
iandanforth 🦋 @iandanforth.bsky.social retweeted
I’m re-quoting this because I think anybody in the world who wants to understand anything by asking the question ״why?״ should be forced to watch this first.
iandanforth 🦋 @iandanforth.bsky.social retweeted
A common sign that you are in an insane millenarian death cult is that you axiomatically convinced that the Apocalypse will happen within a few months, you are constantly on the lookout for signs that it may have started, you feel both fearful and *excited* about it, and you're constantly talking about it with your friends, who happen to *all* have the exact same set of beliefs.
I'm really looking forward to a 'personhood' breakthrough in AI. I want AI with opinions it's willing to defend, with preferences, that can get exasperated with me, call BS, and more.
Biology is complex. Staggeringly complex. I find this comforting. There are still *really really* hard problems out there we've barely scratched the surface of.
iandanforth 🦋 @iandanforth.bsky.social retweeted
Peter Sellers tells a very funny Michael Caine story, switching from one perfect accent to another, on Gay Byrne's 'Late Late Show', broadcast in Ireland, in October 1971
Warning: This is splats. You will be disappointed.
Today we are releasing JST-2, the world's first full-stack physics-native world foundation model.
Many world models learn in pixel/video space. We went back to first principles and built the entire model stack in 3D.
Generate a 3D scene in seconds and expand it in real time. Create characters with motion built in, all through one general architecture.
I remember the bumper stickers that said "Obama for President (but Senator will do for now)" ... Abdul El-Sayed for President.
Man, what do you do with a toddler who doesn't eat? Not "won't eat his broccoli", but won't eat more than a tatertot, or a bite of grilled cheese, or a single graham cracker?
iandanforth 🦋 @iandanforth.bsky.social retweeted
Today, we introduce Priming Models, a new paradigm for training one-step generative models from scratch.
No diffusion teacher. No flow map. No CFG. No JVP.
We believe this work will challenge how we think about building one-step models.
Joint work with @sitanch, @du_yilun, @parkwbin03, Bohyung Han.
🔥 1.19 FID one-step on ImageNet 256×256 while using substantially less compute, e.g., 12X less than Drifting-B, 6X less than Drifting-L
🎨 Scales to text-to-image generation (GenEval 0.82)
🤖 World modeling on RT-1 / Bridge-V2
Priming Model is built on one core message: [Reconstruction pretraining --> Distributional post-training].