Tag
Briefings and ideas about World models from AI podcast discussions and the research they cite, with links to the original sources.
Biohub's head of science Alex Rives argues that cell models are held back by what biology can measure. His examples are whole-zebrafish imaging and a laser phase plate for electron microscopes, so far tested only on isolated proteins.
Biohub's head of science, Alex Rives, says virtual cell models fall short above all because biology can't yet measure what matters. He expects an information-level cell model within about five years.
Josh Gao, CEO of robot-data supplier Mecka AI, says labs train on about a million hours of human data, tens of thousands of gripper hours and a few teleop hours. He argues scaling that alone won't yield a robot that learns any task by watching.
On Moonshots, Salim Ismail argued robots will start in dull, dirty and dangerous work and reach homes later; Peter Diamandis expects them to move past that within a year, if the dexterity and capability the panel discussed are really achieved. Anousheh Ansari said safety comes first and doubted the humanoid form.
Dyna Robotics chief scientist Jason Ma argues that robots should predict future camera video, not only their next moves. He says the bet could fail if touch or sound makes rival models far more data-efficient.
Jason Ma, chief scientist at Dyna Robotics, says robotics has no internet of data. Dyna pretrains on a million hours of human video, adapts each task with minutes to hours of robot data, and Ma expects synthetic environments to add more.
Dyna Robotics' Jason Ma says hour-long robot jobs such as laundry need a planning layer around the low-level action model. He names fragile robot hands, slow hardware testing and vision-only data as limits on more general work.
Goodfire researcher Eric Bigelow resampled a GPT-3.5 model's reasoning at every token and saw answers collapse at single tokens, some arbitrary. He argues the "decision" happens in sampling; newer reasoning models often fork more smoothly.
Google DeepMind's Pushmeet Kohli said scientists need models with calibrated confidence and known limits more than human-readable inner workings. Biohub's Sal Candido said protein models hold knowledge not yet unlocked.
Keerthana Gopalakrishnan of Google DeepMind said robot training will probably blend teleoperation, sensor-equipped handheld rigs and human video, each trading scale for precision. On world models, she said the jury is still out.
Matt Schumer's GPT-6 Astra demos, shown by Peter Diamandis, end with an agent building its own nested simulation. Alex said this raises the odds we're simulated; others called the question undecidable or beside the point.
World Labs' Atlas generates video along a user-set camera path and rebuilds scenes from photos, inventing what was not photographed. On Moonshots, Alex inferred from its documentation that it trains on Gaussian splats.