Hello World! 🌎 At Clone, we are making synthetic humans, starting with a 1:1 copy of the human hand. Here’s the Clone Hand’s first manipulation demonstration, manually preprogrammed in a few hours!
@pokervincent
@wildtiktokss
Growing up, I played Monopoly for PC by the book. It’s a fair and fun and not-nearly-so-long game when you follow the game instructions :)
Pick-and-place with Clone Hands is pretty straightforward — 99.999% accuracy not needed. Here's Clone’s VP of Engineering showing off his new telekinetic powers 🦾👽
Here's the first Clone Hand connected to a UR5 arm!✋x🦾 pick and place is easy for the Clone Hand even with inaccurate tracking software due to its high fidelity biomechanics.
Meet the Clone Hand XVIII! This is our most advanced, anthropomorphic hand yet with 24 DoF, new ligaments and connective tissue, and a quiet and portable hydraulic powering system!
@MoonstruckED__
@pokervincent
@wildtiktokss
The game is more fun with humans, but it sounds like your family doesn’t make up debt covenants and strange trade terms every other turn after hotels have been built lol
Robotics folks saying this is teleop and thereby unimpressive have missed the point. Training the autonomous policy on 15 minutes of teleop demos can be done with relative ease if the hardware isn’t breaking.
My disdain for this design is due to the rigid body that will cause
Lots of engineers want to work at Clone. Some are more serious than others. The Clone Hand inspired this 19 year old student in Korea to build frictionless anatomical joints. He produced incredible results in just 6 months, better than any cover letter.
Clone has a team of 12 and doesn't hire anyone who built motor-actuated robots before. To a neural network, the control input and sensing feedback are just numbers. Motors' perceived edge in linearity / controllability disappears.
Steve Jobs, when he formed his first iPhone team of about a dozen people, told them "you are not allowed to hire anyone who has worked on a phone before."
I didn't really grok that until I saw this.
What "rules" are about to be broken due to AI?
All of them.
A NeRF has
Robotics will be solved and scaled up at Clone. People who are really serious about software (ASI) build their own hardware (androids with human-level hands).
Even with teleop+IL, trying to shoehorn human-level motor abilities into an open-source, 6 DoF snail robot will result
Nice summary of the zeitgeist of robot learning discourse these days: is scaling enough for general-purpose robotics?
I’ll add a perspective: it depends on what you view as “solving” robotics.
#1
99.X% robustness and deployed broadly across society in homes, factories,
Introducing 𝗥𝗼𝗯𝗼𝗣𝗶𝗮𝗻𝗶𝘀𝘁 🎹🤖, a new benchmark for high-dimensional robot control! Solving it requires mastering the piano with two anthropomorphic hands.
This has been one year in the making, and I couldn’t be happier to release it today! Some highlights below:
Clone’s seed round will enable us to manufacture the most powerful and biomimetic androids in the world at a fraction of the cost that most believe to be possible! You can invest over the next 60 days at 👽
Hi Clone Community — Clone’s crowdfunding campaign just went live at !
Check out our pitch for more details about the company and our path to androids, and consider supporting the campaign!
@ErichWeikert
@Bbbn192
@clonerobotics
I see the resemblance! But seems it works like Nebula’s arm in GotG3 :)
Łukasz and I loved the scene where all of injured Nebula’s joints crack back into place like a zombie haha
This guy is a genius who is seeing the future with his systems. First he produced all the best results in unsupervised RL. Then he cracked near-infinite context transformers, which he has now applied to video with
@wilson1yan
👏 Excited to make this large model colossal at Clone!
We are excited to share Large World Model (LWM), a general-purpose 1M context multimodal autoregressive model. It is trained on a large dataset of diverse long videos and books using RingAttention, and can perform language, image, and video understanding and generation.
@DrJimFan
It’s an impressive demo! Still visibly has <half the grip of the Clone Hand, and it’s missing 14 DoF on the Hand, including finger abduction/adduction.
For pen spinning, the Clone fingers will work at least twice as fast :)
@DrJimFan
@clonerobotics
Thanks for the shoutout! Ameca is a HUGE step forward from Sophia. Big fan, but EA had to move the camera from the head to the chest to save skull space for motors, and it’s still missing a few degrees of freedom. With 40 facial muscles, Clone can make the face truly 1:1!
@tunguz
DishBrain’s Pong results were statistically insignificant unfortunately, but it’s still a strong research direction. Here’s the only other lab I know working on this: .
@MelMitchell1
@geoffreyhinton
“Cognitive science” folks are abundant, but there are few human beings in the world like Geoff who devote most waking moments of their entire lives to understand how brains work.
If you want to create an embodied general intelligence and bring the world’s most human-level androids to life, send us a DM or email your CV directly to dhanush
@clonerobotics
.com.
There is a real opportunity to do this right. Sad that Curai is the only AI healthcare startup on this list. That is where great AI will come so sign up to this AI Executive Order! $5/mo digital first primary care and chronic care overseen by human doctors is possible for every
@SmokeAwayyy
@clonerobotics
Not just yet since we are still finalizing actuation and sensing systems over the next few months. But it should be known that the Hand is the hard part. Scaling up the musculoskeletal design to the rest of the body is quite straightforward for us.
But how will the AGI physically kill us??? 🤖💣
@ESYudkowsky
names a couple specific methods:
* Pathogen-aided mind control
* Artificial life forms that reproduce in our atmosphere
These are just human-understandable lower bounds, to help you gain respect for superintelligence.
My first tweet is a flex. I built a magnetoplasmadynamic thruster when I was 13. Here’s a ~1.2 MW discharge with argon propellant operating at 270V and peak currents of 7-10 kA
Here’s one of my favorite scenes from any sci-fi series or film!
———— Spoilers Ahead ————
Context — Mother is an android WMD raising human children and the Quantum Trust is the incorruptible governor of the human colony.
@spakhm
Damn Thiel said the only remaining pro-China constituents in the US as of 2020 are (1) a subset of asset gatherers on Wall Street, (2) universities, and (3) Hollywood
Apple will probably win LLMs.
They'll release a model that runs locally/native on your iPhone, is good enough, has great privacy/security and has access to all your personal data.
The advantages of larger foundation models will be negligible for 99% of tasks.
@ylecun
What other option is there to control these robots in the open, real world, where the environment keeps changing and new adaptations are necessitated?
@silly_silicon_
That would help to collect high quality demonstrations, but we can also collect those demos without a glove and only more accurate tracking. This teleoperation system was hacked up in a week without using any learning.
@DariusPresent
Protoclone will be the bestseller for a while!
There’s no probability. We won’t launch a product in the uncanny valley. Ameca and Mesmer are almost out, but not yet. Their motors in the skull don’t capture all degrees of freedom in the face, but our artificial muscles can do it!
@akber1khan
@joinrepublic
@clonerobotics
Yes, Abher. Another investor made an offer to get us all the resources we need to make low-cost androids, and we wouldn’t have been able to close the capital due to a several month delay in converting the Crowd SAFEs to equity as a part of a priced round. Deals unfortunately
@tobias_sterbak
@ilyasut
The price of energy and the cost to manufacture goods approaches zero in the intelligence limit. Maybe the manufacturing decentralizes further. Then, each human being’s endlessly manufactured and imitated desires can be perpetually quenched!
Prediction: LLMs are going to hit a wall after GPT-5
Reasons:
- No one other than OpenAI has released a GPT-4 class model yet
- We are soon going to hit a data and LLM-generated synthetic data limit
- GPT-4 already uses several RLHF tricks to improve performance
We will have
@yangWao
@0xnullcline
Neuralink’s major innovation, from what I understand, is the neurosurgical robot that can implant electrodes with greater speed and scale.
Making NeuE probes that don’t elicit an immune response for over a year is still an open problem.
@sama
I had an angry tweet ready to go in drafts several hours ago. Crazy how quickly we go from not ever having something to being upset when it’s broken for a day.
@balajis
's network state movement is real. I met my co-founder (sui generis) on YouTube. Obsessed with androids, we started a company together. I moved to Poland, enabling us to work together on the hardware. Now Clone is crowdfunding on Republic and recruiting internationally 🚀
BREAKING: A senior U.S. intelligence official says Russian missiles crossed into NATO member Poland, killing two people.
A Polish government spokesman did not immediately confirm the information, but said leaders were meeting on "crisis situation."
It's easy to find ways to improve when you are genuinely curious about something. Rather than asking yourself, "How can I be better at this?" start by asking, "How can I be more curious about this?"