“We’re predicting the next optimal pedagogical action… sometimes text, sometimes silence, sometimes draw a diagram… difficult for language models”
♫
Said on This Week in Startups
“The model is basically predicting at very high frequency… at least 50-dimensional vector… 10-20 times a second… potentially predicting one to two seconds out”
Latent Space: The AI Engineer Podcast · Dyna-2: General Purpose Physical Agents with Hour-Long Autonomy - Jason Ma, Dyna Robotics
Play from 44:27
44:27
Audio streams from the publisher’s own feed. Verbatim does not host or copy it.
What was described
train a model that outputs a 100-dimensional continuous action vector at 10-20 Hz, predicting 1-2 seconds ahead (≈100 frames)
≈100 action dimensions, 10-20 Hz, 1-2 seconds horizon
Said nearby on the same subject
“The more you post-train models the less diverse their outputs become”
EB
Eric Bigelow
"The Cognitive Revolution" ·
“We trained models on internal data and then predicted on external data… total number of tests … like 13 million across 500 protein targets.”
IQ
Ian Quigley
Latent Space: The AI Engineer Podcast ·
“A billion dollars goes a long way in a country with 100,000 people or a million people… it’s like $1,000 ahead, $10,000 ahead”
♫
Said on Unchained
Moments like this, every Monday
The strongest new moments on your topics, in the speakers' own words. Free for three topics.
No card. No password. One-click unsubscribe.
Track a competitor?
Pro watches up to 25 names or phrases and emails you the morning after they are said.
See Pro, $29 a monthVerbatim quotes a short extract and links to the episode; the episode belongs to its publisher. Remove a show.