We just shipped our biggest @rerundotio open source release ever, and our commercial product Rerun Hub is now available as private preview. I’m deeply proud of what the team has done here and very excited to share more publicly what we’ve been working on for the last year and a half.
We’re building a new data layer for robot learning
@mark_cummins I ported today a 15k codebase from GAS to Go, started around 11 was done by lunch time, 400k tokens burned, across 5 milestones and 37 mins of thinking, in basically one-shot.
There’s little need to debate architecture choices, you can just make it into existence and test.
mosaics were signs of refinement and elite taste, the ancients viewed public surfaces as opportunities to tell stories and spread beauty
then they got abandoned largely due to cost, custom mosaic being $7,000/m²
But robotics can fix this?
Make profits, share profits. Keep things simple. Real money, not fake money based on a fickle secondary or public market. Nearly all options expire worthless.
And let them start earning the real thing quickly, accumulating more units the longer they stay up until a cap at 10 years. Doesn't matter the position or salary — reward longevity (fit), not role.
Sharing profits aligns with a healthy business.
This is how we do it:
https://t.co/bVoHb28gQZ
These aren't small sums. About 1/3 of our company got 6-figure profit shares last year. Real cash they can buy real homes with, real college educations with, real whatever with.
Sevda Arif Kurşunoğlu, J Robert Oppenheimer, and Pakistani Nobel Laureate Abdus Salam converse outdoors in Miami, Florida
Sevda, the widow of Turkish physicist Behram Kurşunoğlu, is now 102 years old and still lives in Florida
(Photo courtesy of AIP)
#Oppenheimer#AbdusSalam
Going from that to this in one lifetime is why many societies around the world still have the fire of progress in their belly - they witnessed it first hand after all, and yet many others have lost it. That forward momentum is a powerful force.
You can feel it as you travel.
@OfficialLoganK
Gemini is awesome.
If you ship memory and give people a way to pry away the context from ChatGPT they've built and load into Gemini many will switch faster
ps: also pls make it possible to archive/delete chats
What if we're approaching robotic manipulation all wrong?
@ilyasut makes a fascinating point here:
Current approaches throw massive amounts of data at the problem — millions of simulation steps, huge compute.
And still, robots can't match human dexterity. Real-world learning of new skills? "Very out of reach."
Meanwhile, humans pick up manipulation tasks almost instantly.
Our dexterity is unmatched. And we do it with remarkable robustness — no reward shaping, no curriculum design, no training instability.
Why the gap?
Evolution. 500M years of optimisation for locomotion, vision, and manipulation.
For these ancient sensorimotor tasks, maybe brute-force data isn't the answer.
Maybe we should be copying the biological algorithms that already solved it.
CC: @dwarkesh_sp
Our model can now learn from its own experience with RL! Our new π*0.6 model can more than double throughput over a base model trained without RL, and can perform real-world tasks: making espresso drinks, folding diverse laundry, and assembling boxes.
More in the thread below.