Our 100th episode! A look back with hosts Andrey, Sharon, and Daniel
15:58 Share

Hello and welcome to Skynet, today's Let's Talk AI podcast where you can hear AI researchers discuss what's going on with AI. I am Andrey Kurenkov, a PhD student at the Stanford Vision and Learning Lab, where I focus on learning algorithms for robotic manipulation.

And I am Dr. Sharon Zhou and I graduated from my PhD from the Stanford Machine Learning Group, advised by Andrew Ng. I do research on generative models and applying machine learning to medicine and climate. I'm Daniel Bashir. I did my bachelor's in CS math at Harvey Mudd College where I did research in information theory and ML. Now I'll be working on language models over at Sama Nova Systems.

And we are all hosts of the Let's Talk AI podcast and are involved in its production.

And this will be a slightly interesting episode. Usually we cover news or interview people. But since this is the 100th episode of this podcast, we wanted to do a bit of a discussion as to what we got out of it so far, looking back at the process and so on. So this will be pretty short. And if you are a listener, maybe you'll find it fun.

And we always want to improve, especially now that we're on number 100. So if you're a longtime loyal listener, we would really appreciate your feedback at bit.ly slash LTA survey. And LTA is just Let's Talk AI. Yep. Also, if you're a newer listener, you know, if you've been liking the show, also, we would appreciate your feedback. And just again, bit.ly LTA survey, and we'll also have it in the description.

And with that plug being over, let's get into a bit of conversation as to this podcast and our involvement. And as our first topic, we are thinking to talk about why each of us wanted to do a podcast in the first place and maybe this type of podcast. So since I'm talking right now, I'll cede it to Sharon or Daniel and then I'll go later.

Yeah. So a huge reason why I started was to stay up to date and hang out with you, Andre. And of course, with the pandemic, it became one of my most social things at the very beginning. And I loved I loved having that. Yeah. And I so far really enjoyed it.

Yeah, I definitely agree with the staying up to date. I think that doing this roughly every week kind of forces me to keep up with the news in a way that I definitely wouldn't otherwise. Besides that, I do have a pretty major addiction to podcasts in general. So I think that the idea of helping out with one seemed like something that I would enjoy.

Yeah, that also, of course, holds for me. I can also think of one other reason. This started after we had the last week in AI newsletter for quite a while. And most of our episodes are these last week in AI type episodes.

And, you know, also with that, I found it was really fun to sort of keep on top of what's going on in news. And I've noticed that I knew much more than I used to. So, yeah, that kind of motivated me to do a podcast. And it also just seemed like fun. So, you know, that also was a motivator. And I think it turned out to actually be pretty fun.

Alrighty. And then, you know, after we started it, I guess let's reflect a bit on our experience of it. You know, what did we enjoy? What were some surprises about it? You know, aside from what we mentioned on the social aspect and coloration, you know, talking into a microphone and, you know, figuring out a format. How was that?

Yeah. So one huge surprise, um, in terms of all the news that we've gotten this past year, year and a half, uh, has been for me at least was the CMS approval, uh, thing in healthcare and having, you know, Medicare, Medicaid really is supporting, uh,

medicine and AI medicine and seeing that trend just grow, especially with COVID as well, just the data sharing that is involved across countries. And so that's been really exciting for me to see.

That's pretty incredible. Yeah. On my side, I guess things that I've followed closely or thought were pretty interesting were a lot of the various attempts at legislation we've seen, I guess, most recently, the EU Draft Artificial Intelligence Act. I think a lot of it has been noticing many of the issues in these acts. So there's been a lot of recent work. I think, Andre, you pointed out one, too, that highlights some of the loopholes and

Basically, an effective portions of things like that, but I think just get to the heart of how difficult it is when people want to try to regulate that sort of technology. So that's been very interesting to follow.

Yeah, and as for me, I guess, you know, there's many individual stories that I found surprising. But on the whole, I've really enjoyed, you know, coming across surprising applications or, you know, hacks of AI. So we've talked about the GPT-free powered NPCs and the hot dog man. That was really fun. Some of these stories of voice acting, you know,

AI in the cosmetics industry. All that was interesting and things that I would not have thought of or expected. Aside from the big ones we know like autonomous driving, there's a lot of these smaller areas and that really made me more conscious as to how pervasive AI probably will be in the future.

One trend I loved has been just generative everything this past year or two, especially in the speech and audio space. I just feel like that space has really exploded and we'll just keep seeing more and more creative uses and hopefully not super malicious uses as well.

I thought that was really exciting. Yeah, I guess too, one of the applications Andre mentioned, the GPT-3 powered NPCs was very cool. One other thing that was pretty interesting that I think we saw too was the revival of dead people, like both dead artists but also people's loved ones through generative AI, which is, I guess that's more on the, I'm not totally sure how I feel about that, but very interesting side as opposed to definitely positive.

But I guess we've definitely seen things on both sides there. It's interesting because it's kind of like modern necromancy. It's like we have it a little bit, this weird way. AI truly is magic, turns out. Or just probability. Yeah. Well, magic is just probability, right? Yeah. Yeah. We already know future where technology is magic, apparently. Mm-hmm.

Also on the trans side, I guess something I was a little surprised by or, you know, noticed is how many news stories we cover are kind of negative and, you know, are a bummer.

I think, you know, especially early on, we noticed it was all this stuff about bias and facial recognition and not so much, you know, the more positive aspects like, you know, nice applications of AI or things that improve the world. And I think over time, we've started to see more towards

You know, you having a real mix of, you know, things like facial recognition, being everywhere, but also these more positive things of how AI is improving world and where it's showing up to make things better.

I feel like it really gets at the sensationalism that we see in the news and that they focus so much on the negative. And so a lot of the stories we do cover are negative. So we have to make a very conscious voice to choose more positive news stories around AI. But that is that's definitely the public sentiment, to say the least.

I was going to say roughly the same thing that I think it's in part just a function of like the broader kind of cultural moment we're in. Having seen, I guess not just specifically about AI, but that kind of increased skepticism towards tech companies and the sort of grand ambitions of changing the world and making people's lives better that they'd laid out. And

And people being increasingly skeptical of that, worried about, you know, what are companies doing with my data, things like that. And so I think a lot of that sentiment, as you pointed out, Sharon, kind of transferred over to these AI systems. And so we saw some of that skepticism manifest as well there. Yeah, I think with regards to skepticism, it's kind of interesting. You know, we were already...

a good few years into AI and especially early on, I think it was even more sensationalist and had more clickbait. And I wonder, yeah, as the years go by and AI gets deployed,

If people sort of start being more aware of it generally, you know, sort of how we became with the internet or computers or other technology, you know, early on coverage was kind of like, oh, wow, this new thing. And over time, maybe coverage will assume readers will know more. For instance, like with this Anthony Bourdain voice cloning story, you

I think the deepfake part of it wasn't as novel because that's kind of an old story at this point. And it was more about voice cloning showing up. So I found that a bit interesting and I'll be curious to see how that develops in the future. Well, trends aside, you know, let's look back maybe at some favorite interviews you've done or, you know, some favorite kind of stories.

things that we found about, uh, developing the podcast. I think looking back, uh, you know, we tried to go for more interviews we found interesting, uh, as opposed to let's say big, big names. And I really enjoyed some of the ones where we, um, interviewed sort of real specialists. So for instance, uh, there was one where we interviewed, uh,

researcher that looked into the deployment of robotics and automation in the workforce and how it impacted jobs. It was very interesting in particular in the context of COVID hearing that actually there wasn't an increase in automation which would be sort of the intuitive thing and in fact you know there was less need of robots and more need for people

So that was really interesting and I found I think most interviews actually quite interesting personally. I really enjoyed the interview with Kate Metz who wrote the book Genius Makers. It was just a fascinating exploration into the history of all of this deep learning stuff.

it was a history that I wasn't you know very familiar with I knew like bits and pieces but not the dramatized version of of course and also not all the little details that he was able to masterfully put into this book and I found that really interesting it was also a really fun and funny interview yeah that book is on my bookshelf and I keep meaning to get around to it

So yeah, that sounds interesting. Daniel, I know you haven't done interviews yet, although you probably will soon. So yeah, maybe you can go a bit more into, I think you were excited to get into producing a podcast and being part of it. And, you know, we've chatted on enough. So yeah, what did you find fun about the process so far? And, you know, what are you looking forward to with respect to interviews maybe? Yeah.

Yeah, I think that the really fun part about it is a lot. I think, you know, what's kind of nice about podcasts in general is just the fact that it's a really kind of nice, useful medium to get through a lot of important ideas, but things like venues into a very digestible format. And so I think going through that process of taking articles, keeping up to date, distilling things and

and presenting it in a reasonable format, but also kind of maybe getting to give my own context or background that maybe the articles didn't contain on top of that was always really enjoyable. I did not do too much of the audio editing except at the beginning. And most of what I've done has been for like,

Basically silencing really small sounds and things like that, which has become, I think, less and less needed because I think one of the weird things that happens over doing podcasts for a while is you notice your own breaths a lot more. So I think that my breath control has become a very big thing as I as I record.

And so I noticed those a lot less, for example. So that's made my job a little bit easier in that regard. But I think just the whole process of distilling things, trying to give that kind of context and background has been fun. Yeah, I guess in regard to interviews, definitely looking forward to getting started on those. I know I have one kind of lined up pretty soon. Yeah, sounds great. And yeah, respect to, you know, getting more skilled. I think it's definitely been interesting for me and Sharon as we discuss new stories every week.

you know, it's funny early on, we were definitely a little bit more awkward, you know, we didn't quite get the flow of it. And then over time, you know, we almost kind of expected, we didn't need to talk about transitions on where we had done on a given topic. We just sort of picked up on it and, you know, naturally went back and forth. So it's been really interesting and it's been pretty fun to sort of,

get into it and get into a state of flow where things just sort of feel natural.

And I will say that feedback has been really useful for us in terms of changing up our style. So not to plug it again, but to plug it again at bit.ly slash LTA survey. We would love to get your feedback because that has actually changed the structure of how each of us speak and also how we structure these episodes at all. And with that, thank you so much for listening to this week's episode of Skynet Today's Let's Talk AI podcast.

Please subscribe to our weekly newsletter at skynettoday.com. Subscribe to us wherever you get your podcasts and don't forget to leave us a rating and review if you like the show. Please be sure to tune in next week. Be sure to tune in next week. Be sure to tune in next week and in the weeks following that. We're excited to keep doing the podcast until we reach 200, 300. We'll see how far we get.