https://nitter.cf/t.co/mtsgLYqwNY The future is robots Prev. Physical Intelligence @physical_int, @Google, @Meta, @Microsoft
Palo Alto, CA
Joined July 2008
- Tweets9.1K
- Following1.3K
- Followers4.2K
- Likes13.8K
Crazy that Google accumulated $100B in cash, then discovered the most lucrative capital-intensive business ever. Being uniquely positioned to scale as the only vertically integrated player in AI datacenters, they…
…did stock buybacks. Then raised a bunch of debt. 🤔
Absolutely insane that Google didn't invest that cash in a huge TPU buildout from 2020-2024. Real lack of confidence in their own business. Too afraid of Wall Street analysts.
Of the top 11 companies in the world, 10 of them were founded on the US west coast. And one in Taiwan.
I see a lot of people "updating" on current events. I stand by my previous statements. The risks of the current generation of AI models were massively exaggerated.
Yeah, some websites are getting hacked. The real world consequences are tiny. The latest one is Binance. They lost billions, but their customers are fine. It hardly affects you at all unless you are an owner of the company. They will continue operating just as they did after the last several non-AI hacks.
This won't always be true. There will be a point where the models are powerful enough to start causing serious real world consequences for large numbers of people. We are still far from that point. My complaint has always been that exaggerating the danger posed by models and applying onerous safety measures prematurely not only slows useful progress in AI and disadvantages us vs. China, but is also actively harmful to the cause of preventing real damage when the models do become legitimately dangerous.
> "Max"
> not the largest
We’re renaming our Max plans:
Max 5x → Max
Max 20x → Ultra
Max is still $100/month, and Ultra is still $200. If you’re a subscriber, there’s nothing you need to do.
We’re also removing the 5-hour limit for verified Ultra accounts, starting today with most existing subscribers. Newer accounts keep a temporary limit until they’re verified, which helps us prevent fraud and abuse. You'll get an email when it's removed.
Weekly limits remain unchanged. View our plans here: claude.com/pricing
World models/learned simulators are the future.
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/
James retweeted
Imagine how frustrating it would be if you prompted your agents super clearly, "Do no work on Saturday at all" and all of a sudden they're reward hacking shit like this
James retweeted
Here are all of @physical_int partners showcased in the video.
@ActorLabs
@almond_robotics
Argon Robotics (CEO @Deepans36819800)
@ChefRobotics
Cloud Chef (CEO @nikhilabm)
@DeftRobo
@dm_robots
@goformic
@FoundryRobotics
Frontier Machines (CEO @bajpayee1990)
@gen_instinct
@intuitioncore
@masa_zeals
@Robotcom
@Ultraroboticsco
@weaverobotics
@yonduai
Curriculum learning will have its day
(1/N) Who says you can't do contact-rich, precise manipulation with sim-to-real RL?
No demonstrations, no retargeting, no tactile sensors, no crazy reward tuning. Just 1x, reactive policies trained with scaled up RL.
Excited to finally release Success-Guided Sampling (SGS)! More info in @octi_zhang's thread and more thoughts of my own below! 🧵
Website here: sgs-rl.github.io
As I keep saying nitter.cf/Darpinian/status/19197…
Imagine telling the Bell Labs guys that in 2026 we would manufacture 4 sextillion transistors and it wouldn't be enough
