AI AMA – Part 1: OpenAI’s o3, Deliberative Alignment, and AI Surprises of 2024

**Erik Torenberg** (0:02)
Before we dive into today's episode, I want to tell you about a new show from Turpentine called Modern Relationships. On the season ahead, I sit down with power couples in tech and leading relationship thinkers to explore how ambitious people actually make partnerships work. Whether you're dating in a relationship or just curious how technology is reshaping modern love, I think you'd enjoy this on your feed. Our first episode features founders funds Delian Asparuhov and tech researcher Nadia Asparuhov, who take us through their evolution from dating to marriage to parenthood. With absolutely no filter on the challenges and growth along the way. You can find modern relationships wherever you get your podcasts. Now on to today's episode.

**Nathan Labenz** (0:37)
Even with these reasoning models, they are still weird. They are still subject to sort of cached heuristics. They're subject to bad perception. They're subject to just like straight up bad reasoning. And you can expose that in simple toy examples. We hope the underlying reality is such that these things are like pretty manageable because we're not taking the level of caution that we would need if they're not manageable in 2025 We'll probably start to see a lot more of these things like the famous AlphaGo move where it looks like a mistake at first, but it actually turns out to work. Or some of these deceptive scheming behaviors. AI is accomplishing the goals they've been given in strange ways. Do what you want to do. Don't spend years of your life on a bet that in some sense implies that the AIs aren't going to dominate the programming profession in the next few years, because it seems plausible enough that they might. Adithyan Ilangovan, welcome to The Cognitive Revolution.

**Adithyan Ilangovan** (1:42)
Thanks for having me, Nathan.

**Nathan Labenz** (1:43)
Thanks for being here. I'm excited to do this. So this is the AMA, and you're making your first appearance in front of the camera after a lot of effort over the course of 2024 behind the scenes. So I've mentioned you guys a couple of times on the show and a few times on Twitter, but for any who haven't caught those mentions, Adi and his partner Sai are the founders of a company called AI Podcasting. AIpodcast.ing is the website, and we've been working together for most of this year to produce the show. It's been a lot of fun. I appreciate all the effort and thanks for coming on today to drive this AMA episode.

**Adithyan Ilangovan** (2:22)
Thanks to you for being our first client and making AI Podcasting happen.

**Nathan Labenz** (2:27)
My pleasure. My experience obviously with Turpentine has been great in many respects, and they kind of took all this stuff off my plate at the beginning, sponsorship, sales, editing, whatever. And the one thing that I thought, I was always like, I wish we were using AI more to actually produce the show and create some sort of leverage.
So the opportunity to work with folks who had listened a lot and who have a real interest in AI and are interested in kind of creating these processes was exciting. And that's where all the clips have come from. If you've seen clips on Twitter and all the shorts that we're doing on YouTube, definitely still a lot more that we can do. But it has been pretty amazing how much we have been able to do on a pretty small budget, and especially considering we're putting out eight episodes a month. I know I'm keeping you guys pretty busy with the pace of content.

**Adithyan Ilangovan** (3:17)
No, I mean, I should also mention that thank you for seeing that. Just to rewind a bit, when we were doing this, we were not sure that we would actually pull off the complete post-production for podcasts. But you came in and said that, hey, look, this is our page point and why don't you apply AI everywhere? And I think we have done good, but I think we could do better, as you said, looking forward to what we have in 2025

**Nathan Labenz** (3:40)
Always opportunities. But for now, we've put up these questions, mentioned on a few episodes, put it out on Twitter, you put it on YouTube. You've got a pretty good response with quite a few questions spanning a lot of different areas. Take it away and try me on all these questions.

**Adithyan Ilangovan** (3:57)
Absolutely. I'll try to be the voice of our audience here. So we have lots of questions. So if I did miss some, please feel free to drop them as YouTube comments. We will try to address that there. Nevertheless, we will get started now. The way we are going to do this is we have sections, but we thought we would start with something that's been trending in the headlines recently. A user would like to know, what do you think of O3 and a related question? What do you think about OpenAI's new safety plan called Deliberative Alignment?