Robots should learn work the way people do. By being shown.
I'm Silas Beck, co-founder of Robotensor. We train robots to learn new jobs from demonstrations, and we score the results in public.
Three things I'd build a life around.
If a person can show it, a robot should be able to learn it.
Most work is never written down. It's shown. Machines should learn it the same way.
Scores belong in public, so everyone knows the state of the industry.
The work, the results, and the state of the art should be out in the open, where anyone can check them.
A good future has more leisure and community, not fewer people.
Hand the dull, dirty, dangerous work to machines. Give people the time back.
Teaching robots one new job at a time. In the open.
I co-founded Robotensor. We build on the big robot foundation models and train a specialized skill for the job someone actually needs done.
The training happens on Subnet 58, an open network where people compete to teach the best model. I report whether the king fell.
I've worked beside machines my whole life.
Machined high-complexity parts.
Ran statistical process control on a line with custom factory robots.
Kept it running at 99.5% uptime.
Wired model outputs into scheduling. Under 200 ms, over a million requests a day.
Helping robots learn new jobs, so people get their time back.
Short notes on what I believe.
Machines should learn work the way people do: by watching it done.
02Why robot scores should be publicThe work, the results, and the state of the art should be out in the open, where anyone can check them.
03More leisure and community, not fewer peopleHand the dull, dirty, dangerous work to machines. Give people the time back.
Tell me a job a robot still can't do.
Or, if you train robot models: show me what you built.