Sonya Huang

Josh Gao says robot training data is still many orders of magnitude behind the video Tesla collected for driving. Sonya Huang asks how much has been collected, and how much still has to be collected.

The field has reached the millions of hours, up from the Ego4D set of 4,000 hours two years earlier. He calls that jump a huge step up. Tesla spent more than a decade on two degrees of freedom, acceleration and direction, with millions of cars on the road and more than a few million hours of video. A robot hand has dozens of degrees of freedom. Relative to that driving record, the robotics field is still many orders of magnitude from where it has to be.

Josh Gao says Mecka puts sensors on people because a robot has to learn how hard to squeeze, and video does not capture that force. Sonya Huang asks why manipulation needs force when self-driving mostly does not. A door handle and an egg are the cases she gives.

A wrist camera can show when a hand touches an object. That view is a rough substitute. It does not measure how many newtons it takes to pull a chair open or twist a bottle cap. Closing the gap means new hardware. Mecka is building gloves people wear so a recording includes how much force a daily task takes.

He calls the broader category sensorizing humans. A phone mounted on someone for the day was only the first pilot, not the wh

Sonya Huang asked about pacing the frontier. One panelist is very much supportive of pacing what happens at the frontier of artificial intelligence in the real world, and of moving as fast as possible in simulation.

In simulation, the science of superintelligence can be developed, along with solutions that help the real world go well and lead to human flourishing. Simulation work should happen as fast as possible so that, when events really happen in the real world, human flourishing has the best possible opportunity. Accelerating artificial intelligence until superintelligence arrives will lead to a really great future for humanity. Things are moving really quickly, and the speed is not so

Sonya Huang asked where the ceiling on simulation sits and when an agent should enter the real world. Sonya Huang said Rich Sutton is very dismissive of simulation. One panelist says the right question is whether a superintelligence can be built in simulation. If a superintelligence can be built in simulation, one panelist believes it will happen 10 times or a hundred times faster, because simulation can be infinitely parallelized.

MetNet is the example. A weather forecasting model from the Google Brain Amsterdam office, built together with Google Research, used machine learning to predict short-term weather forecasts. MetNet was way better than the physics-based simulations. The physics si

David Silver strongly believes the approach behind a super learner will be feasible at some point. The human brain, he says, is the only existence proof of something that has discovered a very general intelligence, works in an embodied world, and learns how to do all kinds of different things.

Sonya Huang asks how confident he is that a super learner will emerge from Ineffable at the broadest level, and whether the nearer goals are a 50-50 bet too. He says great moments in science come down to timing. Whether the approach succeeds or fails is about timing, in the same way a startup has to be at the right time.

David Silver says a super learner has incredible capacity in depth and in breadth. Depth means a system that starts from not knowing anything and discovers things that go way beyond what humans know. People worked out languages, technology, science, and mathematics for themselves. Nobody handed down a blueprint, and nobody handed down a trillion element data set. He says nobody has cracked that in AI yet, and a system that keeps learning and discovering knowledge forever is the essence of super intelligence. Breadth means you can drop the learner into any stream of experience, a simulated world, a digital world, or a physical world, and it keeps learning.

Sonya Huang asks what first contact

Amin Vahdat says serving clusters have to be built because inference demand is too large to cover with older training clusters that are no longer fully used for training. Sonya Huang's picture, that the biggest, newest clusters train the latest frontier model and the rest is recycled into inference, is spot on but not the whole plan.

In a particular year, training might be centralized into a small number of large sites, with network distance kept small, so the sites sit on the same continent or even the same portion of a continent. Other continents can then be left without enough serving capacity, so specialized inference is built across the world. Training clusters will probably be used fo

Bernard Huang is the cofounder of Clearscope. On The Failory Podcast he introduces himself through the failures, not the product. He says he was set on being a founder as a child and started breakdance clubs in middle school. At 31 he calls the previous ten years a long run of trying and missing before he got to this point. Success, he says, does not come overnight for most people, except that when it comes it can come all at once. The path he is summarizing runs through online poker, a Dickey's Barbecue Pit, Mosec, Food by People, and Mushi Labs, and it ends at Clearscope, the SEO product he runs now.