I'm going to post conference talks and other videos as well. Not just podcasts.
I think one of the more interesting bits she touches on is that AI is very good at hard things, but very bad at easy things. The example is that AI is better at chess than the best chess player in the world, but worse at things like basic movement and perception than an 18 month old baby. Another example is that building a robot with an AI system that you drop into a random kitchen and get it to make you a cup of coffee at this point is essentially impossible.
There is a lot of debate around the idea that AI systems are getting much better at intellectual tasks like writing grant proposals, creating legal briefs, finding tumors in medical scans, etc. But that it's really not making any progress on solving the basic tasks that we as humans essentially get for free, again, like being able to make a cup of coffee in a kitchen that we've never been in before, or in the case of an 18 month old baby being able to see a dog once and then recognize a painting of a dog, or a cartoon of a dog.
None of our AI can see a thing once and then identify it later. They all require many data points to be able to make detections later.
A lot of people think that by scaling up these AI systems through faster code and more hardware that we'll eventually crack AGI, but there seems to be a lot of debate about that.