Shaping Science with Nick Dirks
Sep 8, 2026 · 53 min · 12 segments
Hans Peter Brondmo is a technology entrepreneur, former Google Vice President, and a leading thinker on artificial intelligence, robotics, and the future of work. He has…
Hans Peter BrondmoGuest
Nick DirksHost
So, uh, I want to ask you about, you know, how far we've gone since that point in terms of the development of capacity and, and, and robotics, uh, in, in general terms, and also hear your, your particular project today.

But, uh, let's fast-forward, uh, November twenty twenty-two, uh, OpenAI releases ChatGPT.


What, what do you think about, uh, uh, you know-- I mean, on the one hand, LLM's miraculous.

Everybody wants to get in the game, and Google realizes it has to begin to invest in, uh, developing, uh, you know, something that can be, uh, competitive.

But, uh, but in the world of robotics, how, how, how do the, uh, developments in, uh, in, in developing sophisticated language models feed back into robotics? One would think that this would be a breakthrough that would be very directly translatable into some of the work you were doing because it takes reinforcement learning, uh, it takes it to a new level, uh, and it, you know, of course also reaches a point of takeoff.

So, so how are you thinking in that moment of, of, of twenty twenty-two, twenty twenty-three, uh, about the pace of developments in robotics and the pace of developments in LLMs?

But, uh, but in terms of the gestalt of your question, the, the-- We saw the reasoning abilities of models, of language models, as a huge step forward.


A lot of those people worked on our robots, and they don't-- And that's where, for example, we collaborated with them on this Pickup-

... um, the, the bricks, um, project I just mentioned from the beginning.

And so they had been doing s-- They did publish a paper actually called SayCan, and it was one where you could just tell the robot, you were s- you and I were sitting here, and I tip over a Coke can, and I would say to the robot over in the corner, "Hey, robot, I k- tipped, tipped over a Coke can.

Can you please help me clean it up?" You would give that level of abstraction.

So, uh, I want to ask you about, you know, how far we've gone since that point in terms of the development of capacity and, and, and robotics, uh, in, in general terms, and also hear your, your particular project today.

But, uh, let's fast-forward, uh, November twenty twenty-two, uh, OpenAI releases ChatGPT.


What, what do you think about, uh, uh, you know-- I mean, on the one hand, LLM's miraculous.

Everybody wants to get in the game, and Google realizes it has to begin to invest in, uh, developing, uh, you know, something that can be, uh, competitive.

But, uh, but in the world of robotics, how, how, how do the, uh, developments in, uh, in, in developing sophisticated language models feed back into robotics? One would think that this would be a breakthrough that would be very directly translatable into some of the work you were doing because it takes reinforcement learning, uh, it takes it to a new level, uh, and it, you know, of course also reaches a point of takeoff.

So, so how are you thinking in that moment of, of, of twenty twenty-two, twenty twenty-three, uh, about the pace of developments in robotics and the pace of developments in LLMs?

But, uh, but in terms of the gestalt of your question, the, the-- We saw the reasoning abilities of models, of language models, as a huge step forward.


A lot of those people worked on our robots, and they don't-- And that's where, for example, we collaborated with them on this Pickup-

... um, the, the bricks, um, project I just mentioned from the beginning.

And so they had been doing s-- They did publish a paper actually called SayCan, and it was one where you could just tell the robot, you were s- you and I were sitting here, and I tip over a Coke can, and I would say to the robot over in the corner, "Hey, robot, I k- tipped, tipped over a Coke can.

Can you please help me clean it up?" You would give that level of abstraction.
The rest of this transcript — segmented and speaker-labeled, so you land on the exact moment something was said
Search every transcript — by keyword, by phrase, or by meaning, across every show Radar indexes
Trends — what is surging across podcasts, measured against its own baseline
Alerts — when a name you follow appears in a newly indexed episode
No account is needed to search Radar.