Stephen HsuGuest
William Leonard PickardHostYes, if Richard Feynman, John von Neumann and modern AI model walked into your lab, who would ask the best question?

I don't think the AIs are quite at the level of Feynman and von Neumann in terms of asking questions.

So they're pretty good at answering questions where the information is already found somewhere in work that humans have done.

And so there's this last little jump that needs to be made before the AIs are able to autonomously contribute to human research.

And of course, there are many papers on trying to advance the creative aspects of the machines to look at truly novel ideas to take AGI, I think we're already there in many aspects and bordering on superintelligence, to think of novel approaches that have never been conceived by the mind of man.

I think both in software development, which coding, writing software, that's probably the area where machines are most advanced right now, and then areas like math and physics where they're a little bit less advanced.

The key point of discussion among researchers in this area, like myself, has to do with something called taste.

So expert researchers in these fields who are at the frontier have a certain level of taste that currently the models don't really have.

But they're not as good as the best humans in terms of having really good taste for what questions to ask or approaches to problems.
Well, we have to train the machine to somewhat think like us and have our values and our minor creativity compared to it.
Yes, if Richard Feynman, John von Neumann and modern AI model walked into your lab, who would ask the best question?

I don't think the AIs are quite at the level of Feynman and von Neumann in terms of asking questions.

So they're pretty good at answering questions where the information is already found somewhere in work that humans have done.

And so there's this last little jump that needs to be made before the AIs are able to autonomously contribute to human research.

And of course, there are many papers on trying to advance the creative aspects of the machines to look at truly novel ideas to take AGI, I think we're already there in many aspects and bordering on superintelligence, to think of novel approaches that have never been conceived by the mind of man.

I think both in software development, which coding, writing software, that's probably the area where machines are most advanced right now, and then areas like math and physics where they're a little bit less advanced.

The key point of discussion among researchers in this area, like myself, has to do with something called taste.

So expert researchers in these fields who are at the frontier have a certain level of taste that currently the models don't really have.

But they're not as good as the best humans in terms of having really good taste for what questions to ask or approaches to problems.
Well, we have to train the machine to somewhat think like us and have our values and our minor creativity compared to it.
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.