How To ACE AI Product Sense Interviews (OpenAI PM Mock Interview)

1. 0:00 – 2:17

   Intro: Why AI Product Sense Matters

   1. AGAakash Gupta0:00

      There is not a single video on YouTube about AI Product Sense. There are product sense mock interviews, but there is absolutely no information about AI Product Sense. And you know what OpenAI, Anthropic, Google AI, Meta AI, all the top AI companies are asking? AI Product Sense. This is a 45-minute case interview where they will give you a specific problem, and you as a product manager need to, in 45 minutes, speed run through the product management process. So in order to help you ace these interviews, in today's video, we're gonna do a live mock, we're gonna analyze that mock, and we're gonna teach you how to ace product sense yourself. Want help like this? Want mock interview coaching for yourself? Then you will have to join my cohort program at landpmjob.com, where I am taking 30 people to land their dream jobs at companies like OpenAI and Anthropic. And who am I doing it with? Industry experts, including our interviewer for today's episode, Dr. Bart, who has helped over 12,000 PMs land jobs and was a PM at Microsoft, and Prasad Reddy, who spent over 25 years in product management. So Bart and I are gonna get started with this mock. Hey, Bart.

   2. BADr. Bart1:10

      Hey, Aakash. How are you?

   3. AGAakash Gupta1:12

      Awesome. Thanks for having me here at this OpenAI interview.

   4. BADr. Bart1:16

      Don't mention it. Your CV was impressive, and we'd h- we'd love to have you here. So, uh, w- I hope that your journey was, uh, nice, that, um, you had no problems reaching us. If you need any tea or coffee through this interview, just let us know. But other than that, I'd love to just jump to our interview question.

   5. AGAakash Gupta1:40

      All right. Let's do it.

   6. BADr. Bart1:42

      So as you probably know, ChatGPT is lots of things. It's not just chat, but for example, we have an image creation, um, feature that's, well, quite a buzz. However, we want to grow it, and the, mm, quest for you today is to tell me how do we increase the weekly active users of ChatGPT image creation from current 175 mil to 350 million, but

2. 2:17 – 4:10

   The Interview Question Revealed

   1. BADr. Bart2:17

      you have to do it in three months, and unfortunately, we can only spare three engineers for this project. How does this sound?

   2. AGAakash Gupta2:25

      Wow. [laughs] This is why I want to work at OpenAI.

   3. BADr. Bart2:29

      [laughs]

   4. AGAakash Gupta2:29

      This is such an interesting problem, and what an exciting product to work on. I myself am an avid user of ChatGPT image creation. Can I go ahead and share my screen just to make sure we have the right shared understanding about what we're talking about and how it fits in?

   5. BADr. Bart2:46

      By all means.

   6. AGAakash Gupta2:47

      All right. So this is ChatGPT. When you say ChatGPT image creation, I hear you saying it's what we hit after this plus button with Create Image, and it-

   7. BADr. Bart3:04

      Of course, but you can also prompt just to ask ChatGPT to create the image.

   8. AGAakash Gupta3:09

      Okay. You-- So you don't... It doesn't ne- We're not necessarily talking about plus Create Image. We're just talking about ChatGPT outputs an image to people.

   9. BADr. Bart3:18

      Yeah. That's basically the same thing. It just helps the users to navigate to the feature and realize and discover that you can ask for it. We don't really have a, like, an onboarding process that guides you through all the features yet, and we wanted to flesh out at least a little bit while we focus on what matters the most in our prioritization process.

   10. AGAakash Gupta3:42

       Yeah.

   11. BADr. Bart3:43

       And now, after lots of successful releases, it has come to our attention that our image generation can topple the competitors and rock the market, and we hope you can help us with that.

   12. AGAakash Gupta3:57

       So I believe you're-- When you say image generation, that's distinct from video creation, so this is not going to include ChatGPT Sora, which is like the video TikTok app you guys

3. 4:10 – 8:28

   Clarifying the Problem & Scope

   1. AGAakash Gupta4:10

      just released.

   2. BADr. Bart4:11

      No, just still image, uh, that has been there before our video generation.

   3. AGAakash Gupta4:18

      All right. Still images out of ChatGPT. So I think we're aligned. Can I take a minute now just to structure my thoughts and my overall framework?

   4. BADr. Bart4:28

      Even more if you need to.

   5. AGAakash Gupta4:29

      Amazing.

   6. BADr. Bart4:49

      By the way, of course, I'm all here, all ears with you, but if you see me typing or, like, looking away on a piece of paper, I'm just taking notes so I can do a better job later at reporting on this interview. And I'm really excited to hear your answers and see where we can take our image creation from here.

   7. AGAakash Gupta5:12

      Awesome. Just working on my framework here. I'll be ready to present in about a minute.

   8. BADr. Bart5:23

      Take your time, of course.

   9. AGAakash Gupta5:53

      All right. I think I have a high-level structure, and this is just a structure because I really want to do this as a co-thinking exercise, kind of like a whiteboard. So I just created these on a Miro board so we can just talk about them all together. How does that sound?

   10. BADr. Bart6:06

       Awesome.

   11. AGAakash Gupta6:08

       Amazing. So I'm even gonna just prettify this while we're talking a little bit here, put this here, because I really wanna focus on basically six work streams of thought. And so let's see here how we can create some connectors between these things, just to show you exactly what I'm thinking. So I wanna start with the OpenAI mission, vision, and strategy. And my reasoning here, my thought process is we still need to contextualize how image fits into everything, so I think it's worth a little bit of time there. Then I wanna talk about the key users, then I wanna talk about their... W- I wanna solidify, like, a focus. I can't just focus on everybody, but focus on some and look at their problems. Then I wanna brainstorm some creative solutions. Then I wanna look at some metrics and make sure that this really is gonna tie back to our goal of 350 million users. And then of course, guardrails and AI safety and other concerns. What do you think? Does that sound like a good framework?

   12. BADr. Bart7:17

       Well, it's your framework. I would like to see the output at the end.

   13. AGAakash Gupta7:21

       All right, perfect. Let's do it. So as far as I see it, right, you guys are trying to bring AGI safely to the world.

   14. BADr. Bart7:30

       That's the plan.

   15. AGAakash Gupta7:31

       And if we think about image, you know, this is multimodal AGI, right? And so this is gonna be AGI that thinks, that writes, that creates images, creates videos, creates apps and tools too, right? It's, it, it affects the world. And so image is one of those most important things for you guys. And so if I think about, well, how does image gen fit into everything, you know? Image is sort of part of the video. Like, it m- it might be associated. Now, I believe the way you guys have built it under the hood is that you have three separate models. You have ChatGPT, which is your text model. You guys have, uh, DALL-E and improvements to DALL-E, which is your image model, and you have Sora and improvements to Sora, which is your vi- video model. You're actually in the inside, though.

4. 8:28 – 13:11

   Building the Framework Live

   1. AGAakash Gupta8:28

      Is that a correct understanding?

   2. BADr. Bart8:31

      More or less. I will tell you exactly once you sign the, s- sign the NDA, just in case. But-

   3. AGAakash Gupta8:40

      Okay. So let's, let's assume-

   4. BADr. Bart8:41

      But yeah, it's, it's, it's, it's fair to say. Isn't DALL-E Microsoft's, by the way? Like to-

   5. AGAakash Gupta8:47

      Sorry?

   6. BADr. Bart8:47

      ... cut it off. I think DALL-E is act- Microsoft's, not OpenAI's.

   7. AGAakash Gupta8:51

      DALL-E is OpenAI.

   8. BADr. Bart8:53

      Oh, okay. Never mind.

   9. AGAakash Gupta8:55

      Okay. Um, so really under the hood what we're talking about then is we're talking about the DALL-E model, and so that's what I wanted to identify, is that you guys would have an entire research team building the DALL-E model. I imagine that's like a 50-plus research team. When you say 3 engineers, what you're talking about is these are the three application engineers who are sitting on the product side of the team, right?

   10. BADr. Bart9:18

       That's correct. No, no, we don't expect you to improve the model so it generates more, uh, users.

   11. AGAakash Gupta9:23

       Okay. That's interesting.

   12. BADr. Bart9:23

       We just want you to work with what y- w- with what we have right now, or feed, uh, the open, OpenAI image creation scientists with changes that you'd need to sell the product.

   13. AGAakash Gupta9:39

       Perfect. So we do want to think about some things that would influence that 50-person team, because they're obviously gonna be able to move mountains, while we are gonna be able to shape the mountain, build some ski runs, build some lodges on top. And so I'll think about both of those. Just to give you the-

   14. BADr. Bart9:56

       So remember about the timing. We, we do need to be reasonable with our projections.

   15. AGAakash Gupta10:03

       Perfect. So because we only have limited time, I just wanna move on to the next section, unless you think there's anything I missed in the first.

   16. BADr. Bart10:13

       No. No, it looks good. Um-

   17. AGAakash Gupta10:16

       Perfect.

   18. BADr. Bart10:17

       Yeah.

   19. AGAakash Gupta10:17

       So 175 million weekly active users. I think that this is actually, uh, amazing. [laughs] Makes me want to work here. Um, you know, I've worked on some projects, uh, close. So Fortnite, 80 million weekly active users. Um, Apollo, something like one million weekly active users. [laughs] So I worked at some different scales. 85 million is very similar though, I think, to 175 million, and I had some experience when I was working at Google and Google Cloud within the larger ecosystem of Google Workspace at 600 million. So I think I have some idea of what a 600 million user thing is, and what I learned from that experience is that there's a bill- bunch of user groups. And so we're not gonna be able to define everybody. But if we were to think about maybe a, a simple bucketing, right, of the 175 million, let me know if this sounds right. This is just based on what I would guess, right? Probably, like, 50 million of the hardcore, like, early adopters who know all the tech, you know, the AI guys, AI bros. [laughs]

   20. BADr. Bart11:30

       Mm-hmm.

   21. AGAakash Gupta11:30

       Probably like, you know, 100 million are just the people who, like, ad- are technology forward, so maybe they're a little bit younger. Or if they're not younger, you know, they're young at heart and they try new technology. So that's like 150 million. And then 25 million, we probably have, like, the really low tech literate folks who barely understand AI, but they have some use case for it, or they used ChatGPT and they stumbled on it. So I'm guessing this is what the users look like today. Do you think this is roughly accurate? Anything I missed?

   22. BADr. Bart12:07

       Hard to say. We didn't look in depth for images here. Uh, butIt, you'd be ob-obviously welcome as a product manager to investigate that and confirm whether this hypothesis is correct. But let's roll with it. I understand this is a [chuckles] 45-minute interview, and we have to base our numbers on something, and guesstimation here is welcomed. By the way, do you know how many, uh, weekly active users does ChatGPT have?

   23. AGAakash Gupta12:38

       I believe it has 800 million, right?

   24. BADr. Bart12:41

       That's correct. Yeah. So, so maybe it will help you to learn that, well, 20% of them will generate images throughout that week.

   25. AGAakash Gupta12:50

       That's a great point. So it's like 25% of users. So now I wanna think about, like, the goal, right? We're trying to get to 350 million users.

   26. BADr. Bart12:59

       Mm-hmm.

   27. AGAakash Gupta12:59

       Like, where are those users gonna come from? I think we basically already hit the 50 million early adopters, which is why I wanted to point that out, right?

   28. BADr. Bart13:08

       Mm-hmm.

   29. AGAakash Gupta13:08

       They're just gonna be the 50 million.

5. 13:11 – 16:06

   User Segmentation (175M → 350M)

   1. AGAakash Gupta13:11

      So really, when we're gonna add these 175 million, they're gonna be from the tech forward. But how many tech forward really are there in the world? I would say maybe 5% of the world's population, so maybe 400 million people. We know that already 10% of the world's population, a little more than 10%, 15% of the world's population is using ChatGPT. So if we say 400 times .15, right? So 400 times .15, just doing the math in my head, it's like, uh, I'm gonna do it on the calculator so I don't mess it up.

   2. BADr. Bart13:46

      [chuckles] Go ahead.

   3. AGAakash Gupta13:46

      60 million.

   4. BADr. Bart13:47

      It's 60 million.

   5. AGAakash Gupta13:47

      This is not a math exam. So just based on my math exercise, we probably already have the younger, the tech forward people too. Um, maybe some of them are only using it, like, once a month or twice a month or something, so we could create them as monthly a- weekly active users. So I'm gonna go ahead and say, like, maybe we increase that by 50 million, so that's 200 million. The whole point of this exercise was to show that the big lift is really gonna have to be from the low tech literate folks. And I think based on this breakdown right here, it's obvious to us where we should focus for this next area of the framework, which is the low tech literate user group. Does that make sense to you?

   6. BADr. Bart14:32

      Very logical. Thank you.

   7. AGAakash Gupta14:34

      Okay, perfect. So now we need to think about what are the various user problems, and I think I would probab- I'm already, like, just brainstorming on this. I wanna brainstorm these user problems into sort of a different, a couple different buckets, like UI, um, like functionality of app, and the other bucket I wanna focus on is the research model, 'cause we talked about influencing those groups as well. Does that sound right to you as three buckets?

   8. BADr. Bart15:19

      It does, yeah.

   9. AGAakash Gupta15:21

      Perfect. So I won't further explain what each of those is. Um, so in each of these, let's go step by step. So in the user problems UI, and I just love these Miros to look nice, so I'm just gonna take a second to make this look nice. Sorry, I just... I almost think, like, if I structure my mind better in this mind map, I think better. Okay. So the UI of these user problems. So I think, like, one thing as we just showed when I was demoing the feature itself is, like, you either accidentally encounter it

6. 16:06 – 21:01

   Identifying User Problems

   1. AGAakash Gupta16:06

      in a prompt like you explained, or you hit the plus sign. So it's like, it's not an easy entry point into image generation. There is no, um, app like you guys have for Sora, you know? So it makes-

   2. BADr. Bart16:24

      Mm-hmm

   3. AGAakash Gupta16:24

      ... it super clear to everybody, "Hey, this is what you do here," you know? Um-

   4. BADr. Bart16:29

      Well, we are ChatGPT, not ImageGPT.

   5. AGAakash Gupta16:32

      Yeah, exactly. And you guys are ChatGPT, not ImageGPT, right? And so I think that I'm gonna need to take a little more space for my user problems. I'm just gonna expand these a little bit. Okay. So there's no app like Sora. Um, let me, let's just look at the UI together, right? What ends up happening, let's just go through the process together so that we can look at it, you know, an end-to-end UI evaluation here. So, um, generate an image of me speaking on stage. Match me exactly so I can use it as a real photo. So I think, like, this is one of my personal use cases, so that's why I wanted to start with something that I feel we could truly show this is authentic. So I've put this in. Let's go ahead and look at the end-to-end flow. Um, I did hit that Create Image.

   6. BADr. Bart17:44

      Mm-hmm.

   7. AGAakash Gupta17:45

      So it looks like ChatGPT is gonna think. If I hit Think, personally, I'm worried that I might skip to see that thinking, but let's click it anyways. Okay. I took the risk. [chuckles] The thinking is not showing. It has this timer, um, which seems longer personally than what I've personally experienced in my mind. Because I thought we entered the prompt, it's only been 20 seconds, so showing one minute 40 seconds seems high. So I'm actually gonna list that in there. High time counter. Um, afraid to see thinking.It's honestly, it's taking a long time. So I'm also gonna put that in here. Speed. We know that that's on the research model side. You know, I find that this is the best way to learn about a product, honestly, is to just use it and watch people use it and look at the data of people use it and read about how people are using it in reviews and talk to them about their usage in a mom test sort of way. So we're just simulating that because we only have a forty-five minute interview here. We'll just do one or two, you know? So I think this is taking forever, and, um, it's interesting. At least we have some UI, like we show the high level and we're clarifying it, so I think that's nice. But I think like some sort of expectation probably, you know, for folks. And like it's, the counter stopped at thought for twenty-eight seconds, but, uh, it's still working. So high time counter plus resets to lower. So I think there's a lot of UI issues that we're noticing already. And I wonder if it's about the-

   8. BADr. Bart19:27

      Yeah. I wonder if you ever tried other, uh, image AI generate, generating solutions?

   9. AGAakash Gupta19:32

      Yeah. I've tried many of them, and we'll actually compare to some of those in a second. To be honest, I feel like as OpenAI, you guys shouldn't be too focused on the other image gen. But since you brought it up, let's talk about it for a minute. Um, let's talk about it. Let's add it to the framework. So with competitors, what I would say, like the most important things that we need to mentally think about with competitors are like, what technology, what usage, who's winning, and why, right? And so I believe that the competitors that matter the most for you guys right now, we're talking about Midjourney, we're talking about NanoBanana, we're talking about avatar companies like Synthesia. Um, in my mind, these are the top three competitors. Is there anyone I missed?

   10. BADr. Bart20:35

       Hmm. No, I don't think so. That's, that's even more than I expected to see, uh, honestly.

   11. AGAakash Gupta20:43

       Perfect. So let me just create a quick section on each of these. Since we're talking about it, I wanna just, I wanna be holistic, you know? And, um, so we said Midjourney, we said NanoBanana, and we said Synthesia. So let's just quickly

7. 21:01 – 24:44

   Competitor Analysis (Midjourney, Runway, Synthesia)

   1. AGAakash Gupta21:01

      make a note about what we're gonna learn from each of these. So Midjourney, I think the thing is like their images just look way better. Artists like them. They have way more styles. Um, they're, they're ahead of us in terms of the core model. And so bringing it back to our ne- our next area in the framework, right? The next area that we had discussed was the model up here in user problems, and so we're gonna add that, like, MJ-like quality and style. And so that's just gonna be my short form way of saying like, "Big work on the model. I know this fifty-person team is already thinking about it."

   2. BADr. Bart21:46

      [chuckles]

   3. AGAakash Gupta21:47

      Um, so I don't know how I'm gonna add a lot of value, especially in forty-five minutes to the topic since they're already thinking about it, but we've written it down. NanoBanana, you know, I think one of the coolest things is like their image selection and editing and text editing that they have. Those are all really good features. And you have-

   4. BADr. Bart22:07

      Could you zoom in a bit? It's hard to read it at the moment.

   5. AGAakash Gupta22:11

      Yeah. Thank you so much for asking about competitors, honestly. I think you brought up a really interesting area-

   6. BADr. Bart22:15

      Mm-hmm

   7. AGAakash Gupta22:15

      ... for the framework. And then, you know, just this avatar idea. Like, we don't even have avatars at OpenAI, and, um, they look so good. They look real. I think once this ChatGPT comes, yep, it, it doesn't look real. You know? It doesn't train on me. It doesn't look like me. And so that's a really important point. Um, and that goes into functionality of app. So basically, creating, um, a rag training set to look like the person is a really interesting user problem we don't have. I've just jumped into the solution. Let me go back to the problem. Doesn't look like me. Um, doesn't have like a selection, doesn't have editing, image or text editing, right? So I'm so glad you asked about the competitors 'cause it just helped us breeze through the top user problems. So now we can head into solutions if that sounds good to you, because I know we're, we're getting tight on time.

   8. BADr. Bart23:12

      Mm-hmm.

   9. AGAakash Gupta23:13

      Awesome.

   10. BADr. Bart23:13

       One actually question I have. Uh, how would you determine what problems are users trying to solve with image, generative image generation?

   11. AGAakash Gupta23:28

       Yeah. That's a-- I'm, I'm so glad you asked that. Um, let me take a second to think about that after I do this reformatting that I'm doing-

   12. BADr. Bart23:35

       Mm-hmm

   13. AGAakash Gupta23:35

       ... over here. We need to g- we need to give that its proper time to think it through. Um, so let me just repeat back the question 'cause I was busy editing here. You asked, uh, what, how would I know what problems users are trying to solve with image generation, right?

   14. BADr. Bart23:56

       That's correct.

   15. AGAakash Gupta23:57

       Okay. So let's put that here as another bucket so we can talk about it. What user solving? Um, so I mean, the main thing we do is we look at their prompts, right? [chuckles] And I think you guys built this amazing chart which showed how people use ChatGPT. It's the same thing, right? It's a chart of how people use image generation, you know. How many people are generating anime? How many people are generating people speaking on stage? How many people are generating infographics for their LinkedIn posts?Those are just three of the use cases. There are 97 more, I bet. [chuckles] So what are the 100 use cases? Looking at the data of the prompts, and then I think what we should also probably do is we should look at the thumbs up, thumbs down data by the 100 category.

8. 24:44 – 26:25

   What Problems Are Users Solving?

   1. AGAakash Gupta24:44

      So then we could quantify, oh, these are the top areas that have the most thumbs down. We could give that feedback back to the research team.

   2. BADr. Bart24:53

      Awesome. Thank you.

   3. AGAakash Gupta24:55

      Okay, great. So I'm just gonna... I'm, I'm actually gonna put that in here, you know, like for our solutions that we, that we'll talk about. So what I wanna do now, because we're running, you know, against the clock throughout this interview, is, is quickly brainstorm some solutions against each of these problems that we identified, right?

   4. BADr. Bart25:15

      Mm-hmm.

   5. AGAakash Gupta25:16

      And so accidentally encounter or you hit the plus sign, right? Like, what if we just, we blew up the ChatGPT homepage somehow. We added, you know, image [laughs] or some big way. Basically like, you know, design UX, image prominence on homepage. No app like Sora. What if we created an app just for images? An Instagram competitor maybe instead of a TikTok competitor. High time counter. What if we fixed the time counter? [laughs] Um, and afraid to see thinking. What if we improved the thinking UX? Basically, you know, I worked in gaming, and so a lot of my product management principles come from a gaming background. In gaming, we really thought a lot about the load screen, and if we think about this load screen here, it wasn't what we talked about in gaming, fun, rewarding, and headed to your goal. And so thinking about those principles and applying those to the thinking UX. I mentioned I worked at Google Cloud on Google Workspace. What we learned is that these default screens that show up,

9. 26:25 – 29:29

   Solution Brainstorming

   1. AGAakash Gupta26:25

      what you do is you map out the amount of time people are spending in the app by screen. [chuckles] So just like you guys map out ChatGPT usage, and I personally like a horizontal stacked bar chart, and you immediately can just see, oh, everybody's spending time on the loading screen, and so really improving the UX there. Does that make sense?

   2. BADr. Bart26:42

      It does, of course.

   3. AGAakash Gupta26:44

      All right. So functionality of the app. So I think I'm gonna end up loving the solutions in this area just as I'm brainstorming now, you know. What if we had, you know, like I was saying earlier when I jumped to the solution and caught myself, you know, like a rag database to make it look like you. I think this would go-- We already know tools like Photo AI, Interior AI, Secta are blowing up in this space, so I think that would be huge. Another one could be like, uh, a face adjuster or iterator where we continue to provide in, um, it keeps, we keep providing more and more images over time, or the ability to take in like 15 to 20 inputs. It doesn't have selection, so creating selection. It doesn't have image or text editing, so adding image or text editing. I think these are all amazing. And even within each of those, I could brainstorm creative solutions, and so I'm gonna do that. I'm gonna do solution specification as another area for our framework that we can spend two or three minutes on since it looks like we'll have time. Does that sound good?

   4. BADr. Bart27:59

      Go ahead.

   5. AGAakash Gupta28:00

      Awesome. So I'm just gonna add that into our framework here. How's everything progressing so far? Are you, are you tracking? Do you like where we're going with this case? Are we using time well?

   6. BADr. Bart28:13

      Hmm. Honestly, by now I'd be expecting you to be more in a prioritization mode and figuring out what could be done in three months. But we still have time, so that's just how I would focus on that. So don't, don't, don't, don't feel stressed about this feedback. And I love your journey. It's very detailed. You put lots of details into it, and the product management process is outstanding, so go ahead. I mean, I, I'm waiting to be, uh, for the cherry on top of this mind-blowing process.

   7. AGAakash Gupta28:55

      Thank you. I appreciate that. So actually, you're right. We should be heading to prioritization. So let me be quick on these solutions, and I added prioritization to that next element of the framework. Really appreciate your partnership on this. So research model, right? We, we need to have multiple versions, uh, high, slow, fast, you know, amazing quality-

   8. BADr. Bart29:16

      Mm-hmm

   9. AGAakash Gupta29:16

      ... okay quality. We need to just improve the speed altogether, right? Um, we talked about getting the, uh, chart of thumbs down data to the team, and then MJ like

10. 29:29 – 33:39

    Prioritization Framework

    1. AGAakash Gupta29:29

       quality. Again, like showing them inspo of like different styles. And I think using the data of these are the styles that we're doing the worst on. Maybe it's photorealistic, maybe it's anime, whatever it is. You know, based on the Ghibli success, it's probably not anime, but let's look at the data.

    2. BADr. Bart29:45

       [chuckles]

    3. AGAakash Gupta29:45

       And then what users are seeking, right? And so, uh, I think we could also use this data, and just brainstorming off the top of my head, I think the infographic space, the thumbnail space, like for my personal, the, the u- it's not there yet at all. And so those would be two areas where people have needs, and we haven't built functionality. So those are all the potential solutions. Let's go ahead and get into prioritization, shall we?

    4. BADr. Bart30:12

       Go ahead.

    5. AGAakash Gupta30:13

       All right. And I, I'm just looking at my own framework, and I feel like I might need a little more space here, so I'm gonna create a little more space. Um, let's create space by moving these ones down and creating this one a little bigger. All right, perfect. So we are trying to get toAdding in 150 million really to- low tech literate users with 3 engineers with 3 months. These engineers are gonna be using ChatGPT Codex, so they are-- I'm gonna assume that they're really good at doing simple front-end changes, and even in, like, a week, they could knock out every single front-end change that we have here. So assuming that's the case, that we have really good engineers on the team, let's do all the UI changes, and let's size that, right? So because this is just gonna be, like, a week, and I think this could easily generate, like, something like 50 million of the users we need to drive. Then we need to dri- send many more. So if we look at the sizing of some of these, make it look like you and face adjuster iterator, I only think those maybe drive, like, 10 million users. So look like you and iterator maybe drive 10 million users, but they're gonna take time for engineers. You know? It might take a month. So I'm already thinking, ah, that's probably not gonna make it. Like, it's not a no-brainer like the other one where I just prioritized it even without comparing the work and size of the other ones. The idea of improving, creating an app, so again, this is gonna take tons of work. It-- Because we already have so many other apps, I don't know that how many incremental users, maybe 25 million incremental users, and it might take all 3 months. So that's definitely not gonna make it to the prioritization framework. Um-

    6. BADr. Bart32:10

       Just like a s- a, a quick side question. If you had to cr-create that app within three weeks, what would you do? Uh, sorry, three months.

    7. AGAakash Gupta32:20

       Oh, I love this question. Thank you for asking. Okay, let's think about it. Let's just put it over here, shall we?

    8. BADr. Bart32:26

       Mm-hmm.

    9. AGAakash Gupta32:27

       Um, creating an app in three months. That's such a good curveball you've thrown me.

    10. BADr. Bart32:35

        [chuckles]

    11. AGAakash Gupta32:36

        All right. Well, we already had inspiration, right, which is, like, Instagram, right? So we're already, we're going after our TikTok killer with Sora, so let's kill Instagram too, right? And so if we think about Instagram, like, a p- a big part of Instagram is authenticity. And so I would very much prioritize, like, how do we create the best image editor from AI? Because we really do want authentic photos. You know, we don't want AI-generated photos to be the whole feed.

    12. BADr. Bart33:12

        Mm-hmm.

    13. AGAakash Gupta33:12

        Whereas in Sora, that works. So we're really focused on authentic photos, and we want to create AI photo editing. You know, one of the things I personally love is, like, I want to, like, zoom out my photo. Like, AI could do an amazing job doing background fill, like Adobe Photoshop has shown. Or I want to zoom in my photo and increase the resolution, or I want to increase the resolution. I know my friend Bart Jaworski, he often has to do that for his memes. So I think that

11. 33:39 – 36:05

    CURVEBALL: Creating an Instagram Killer in 3 Months

    1. AGAakash Gupta33:39

       creating-- I'd work with the research team to pull together amazing photo editing features, just like Instagram. Maybe think AI-driven filters. And I would create an amazing MVP use case around one core user group, which would be, I think, women who want to authentically show themselves around the world, um, and allow them to just edit that 100x better than Instagram. And I can imagine this app being only three months of work because we're just focused on this narrow user group, this narrow functionality, and I can imagine it growing virally because they're sharing it with their friends. Like, "Look how much more beautiful I look. Look how much more resolution I have. Look how it put on a new dress." And I think that could be huge.

    2. BADr. Bart34:28

       All right. Thank you. Let's go back to your main stream of thought. Sorry for that distraction.

    3. AGAakash Gupta34:33

       No problem. I love that. I mean, now that I went... You know how it is, right? When you spec out a solution, you begin to love it.

    4. BADr. Bart34:40

       Mm-hmm.

    5. AGAakash Gupta34:40

       But based on our old prioritization framework, it's at the bottom. So let's throw it at the bottom. Let's keep moving forward. Let's be objective here. I kind of sold myself on my old idea for a second there. Um, so, uh, the, uh, next thing we had talked about is the ability to select an image edit. I think this is huge, right? Working on editing features, I think this is, this is what's gonna drive everything. This is what's been driving Nano Banana. So I think this is, like, of the gap, this is probably 100 million users, and I do think this will take a long time, like 1.5 months. So we are running out of time. If we look at it, what other features have we come up with? Um, s- multiple versions. You know, the research team has probably already created multiple versions, so it's more about productionizing everything. So it'll probably take us, you know, three weeks. I don't know how many incremental users that's gonna drive us, so that's gonna go down here. It's kind of low. Like, let's do, like, above the line and below the line, you know, because it's a fast interview, so above the line and below the line. Now we need to get the full bridge to 350 million users. And so now we're getting kind of, we're kind of merging into what I had thought of, of, for this metrics section. So actually, what I'm gonna do here is I'm gonna create this

12. 36:05 – 40:03

    Final Prioritization & Metrics

    1. AGAakash Gupta36:05

       as solution specification, and I'm gonna create this as prioritization and metrics, just to try to be responsive to what you were saying. Does that sound good?

    2. BADr. Bart36:13

       Sounds good.

    3. AGAakash Gupta36:14

       All right. And then it looks like I need to make this a little bigger, so we are spending a lot of time formatting this mural. Probably not the ideal thing to do, but it just make, it makes it so much more fun. [laughs] All right. So above the line, we're, we're trying to see if are there ab- are there above the lines, and I think so, right? So what users are seeking, like, this idea of, like, infographics and thumbnails, I think building in those featuresBasically an infographic editor and a thumbnail editor. And let's make this more of like a thumbnail and profile and meme editor. So each of these would probably take us like three weeks, but they could generate, you know, 25 million-ish users. A lot more users than the multiple versions. So where are we at? 25, 50, 150, 200 million users. So I think we've, we needed to bridge 175 million. We have some initiatives that could roughly take us 200 million. We have six weeks. Um, we have one and a half months. So this is about accomplishable in three engineers, three months. So this would be my speed run [laughs] 45-minute prioritization. Does that sound good or do you wanna go deeper? Do you want me to get more rigorous here?

    4. BADr. Bart37:31

       Since we are slowly runn- running out of time and I wanted to leave some space for your questions, I think that's good enough. I mean, it's not really about the features, it's about the journey in those, uh, questions. So, well, I think we sort of hit the finish line, wouldn't you agree?

    5. AGAakash Gupta37:50

       Awesome. Well, let me just take two minutes then. If we think, if you think we've hit the finish line, let me just take two minutes and just finish off my framework, and then we'll go.

    6. BADr. Bart37:58

       Mm-hmm.

    7. AGAakash Gupta37:58

       Sound good?

    8. BADr. Bart38:00

       Go ahead.

    9. AGAakash Gupta38:00

       Perfect. Awesome. So solution specifications. So the UI changes, just to s-summarize, right? We're gonna, we're gonna change the ChatGPT homepage. Oh my God, that's gonna be a lot of coordination for me as a PM, but not that much-

    10. BADr. Bart38:16

        [laughs]

    11. AGAakash Gupta38:17

        ... as an engineer.

    12. BADr. Bart38:17

        We are all one big family here. I can assure you it won't be a problem.

    13. AGAakash Gupta38:22

        Amazing. [laughs] Love that. Hopefully Sam can help us like corral everybody and make some those choices. And then, uh, we're gonna fix the timing calendar.

    14. BADr. Bart38:30

        Don't tell Sam. He, he loves it the way it is.

    15. AGAakash Gupta38:33

        Oh, oh, okay. [laughs] Thank you. We're gonna improve the thinking UX, right? These are all straightforward. Like when I demoed the thinking, if you remember, it had started like one minute 30 second, but then it finalized at 38 seconds. So that's what I mean when I say fix all that. We're gonna add editing features. So what I mean when I say editing features is like in an image, you could select a part of an image, and you could change the color, or you could change some frequency, or you could lasso it. So we're gonna-- That's what I mean, like basic Canva-like editing, not Photoshop level.

    16. BADr. Bart39:07

        Mm-hmm.

    17. AGAakash Gupta39:08

        When I say infographic editor, I basically mean with infographics, what w-we're really doing with the AI is we're putting together text boxes and multiple AI images. And so it's like creating a canvas where they can generate multiple AI images, and they can edit the text. Does that make sense?

    18. BADr. Bart39:26

        It does.

    19. AGAakash Gupta39:27

        And then that's actually the same idea for the meme editor, but in a meme context. So there's this thing called dingboard. We would kind of be similar to how dingboard is created. So that's my sum- thump- my solution specification. I'm happy to go into more detail, but since you said we're almost at the finish line and we're almost out of time, I just really quickly want to mention-

    20. BADr. Bart39:47

        Let me just challenge you on one thing, which, which does, I don't want to say bother me, but I want to hear your, your reaction. Do you really believe that, um, the way we present thing right now, um,

13. 40:03 – 41:16

    Solution Specifications

    1. BADr. Bart40:03

       kills the conversion for the future? Do, do you believe that thinking, though, th- that fixing those UI issues like long thinking times and arguably confusing flow on what happens when you generate an image, does hurt the conversion at, to, to a point where in the next three months, fixing it would, could give us additional millions of users?

    2. AGAakash Gupta40:31

       I do, yeah. Um-

    3. BADr. Bart40:32

       Mm-hmm

    4. AGAakash Gupta40:32

       ... and that goes back to that stacked horizontal bar chart we were talking about. The time you spend on the homepage is actually small. The time you spend prompting is actually small. 99% of the time is spent on that loading experience. And so from my gaming background, from my background at Apollo.io, where we doubled activation conversion rate, I found this is the most important lever.

    5. BADr. Bart40:58

       So you could probably, mm, split that into features that onboard the user and then give him that product delight, to quote one of the books that was recently deli- uh, released, in order to, well, convert them to

14. 41:16 – 43:40

    Safety & Copyright Considerations

    1. BADr. Bart41:16

       weekly active users and not people that were interested one time in image generation.

    2. AGAakash Gupta41:22

       Yes, exactly. And I actually had Nesreen on the podcast. I love that you mentioned delight. I think that product delight is what we're going for here.

    3. BADr. Bart41:30

       Mm-hmm.

    4. AGAakash Gupta41:30

       Awesome.

    5. BADr. Bart41:31

       Well, I do prefer next gen product management by certain Dr. Bart, but like [laughs] I love the concept of like going down to small, intricate details and like making it, uh, an important part of product management rather than overseen, uh, detail. But let's still focus on you, and actually we still, we only have two minutes left, uh, on my counter, and I have a hard stop, as usually it is in those, in our big OpenAI family. So perhaps at, at least I can answer one of your questions.

    6. AGAakash Gupta42:09

       All right. Let me just, even before I get to the question, I just want to make sure, like we're trying to create safe AI.

    7. BADr. Bart42:15

       Mm-hmm.

    8. AGAakash Gupta42:15

       We need to make sure that like people aren't editing images and likeness of third party, right? So we need to create a verification when we build that feature. We need to make sure that, uh, people aren't creating, you know, pornographic or dangerous images.

    9. BADr. Bart42:31

       Mm-hmm.

    10. AGAakash Gupta42:31

        I know it's, it's bad to talk about, but we have to talk about it. Um, you know, like any sort of abusive language or memes or dangerous memes. And so we need to do a lot of red teaming and safety around this feature. I just wanted to mention that.

    11. BADr. Bart42:45

        Mm-hmm.

    12. AGAakash Gupta42:45

        Um, and then shall we go to my question?

    13. BADr. Bart42:49

        Just one quick, quick question to follow up. You did mention Ghibli several times, which gave us a lot of, um, traffic, but at the same time it also put us in hot waters with the Ghibli studio itself. Do you think we should still be like, mm, uh, what's the phrase? Um, you, you can edit my thinking. It's late for me. Uh, how do you say? Like say, like be bold now and say sorry later, or should we already care about, uh, the copyright at this stage?

    14. AGAakash Gupta43:33

        Yeah. Um, I think we should talk about this with Sam because

15. 43:40 – 45:44

    My Questions for the Interviewer

    1. AGAakash Gupta43:40

       the approach he took with Sora, you know, not being so copyright-friendly, going viral-

    2. BADr. Bart43:45

       Mm-hmm

    3. AGAakash Gupta43:45

       ... but then being more copyright restrictive, I think it worked, but people didn't like it. And so we need to get direction from the leadership, but my personal point of view would be, be more copyright safe, um, and not have the part, because if you look at Sora App Store reviews, they're really low. And so my signal based on that data is let's be copyright safe from the beginning.

    4. BADr. Bart44:08

       Love that you both have an opinion and be a team player. Thank you. So your question quickly, and I really need to go to my next call.

    5. AGAakash Gupta44:16

       All right. Thanks. All right, YouTube watchers, that concludes the AI product sense mock interview. Let's go to Bart. Bart, how did I do? What did I do well? What did I do-- What could I have done better?

    6. BADr. Bart44:33

       I only like disagree with like certain small choices which I wouldn't make, but what was there was awesome. There was a clear journey of, from like the vision to the problem, to the solutions with like all the elements of discovery that told me that you are excellent at product discovery without telling me that you're excellent, uh, at product discovery. You had the right amount of guesstifica- guesstimation, sorry, um, on the way. I love that you had like I-- as you said, I threw a few curve balls, uh, at your way to, to make sure that you are not always super steady, but you always had a good answer and then came back to your train of thought without losing the, um, the focus. And that's really, maybe I would like be a little bit more of a team player, just like, like those little details saying that I think that this is how much we'll spend on it, but

16. 45:44 – 50:03

    Interview Feedback & Breakdown

    1. BADr. Bart45:44

       of course we, I would have to at least spend, uh, a little bit time with the team to make sure that those estimations are correctly. Maybe I would also leave a little space to do like inside, um, company research to find out what do we know already about image generation. Maybe we have something that's left on the shelf that wasn't released for whatever reason, or like a banger researched, uh, ticket that's there in the backlog and was deprioritized. But again, this is what I would do. And when it comes to A to Z product management process, which you need to demonstrate on those job interviews, that was awesome. And I would personally probably not do as good as you did because I wouldn't, um, do the Miro board, which is an awesome learning for me to make it visual, to like counteract thing that I struggled in the past on making a narrative that p- people can easily follow, and you just like crushed it with the story that's easy to understand from A to Z. And whether the actual product manager interviewer in OpenAI will agree with certain details, I wouldn't say that those are important. You have 45 minutes to give the person on the other side a tour of your thinking process, that whatever you do, you will come up with the right solutions. And Akash, you demonstrated that beyond any doubt. So like s- maybe A-minus for like personal nitpicks and different decisions, but doing A-plus is, or A would probably require some, uh, Professor Xavier level of mind reading, and I guess I probably wouldn't give more than A-minus to any candidate, just because there will be some disagreement, and working in OpenAI probably know stuff you don't do. But having said that, uh, incredible that you clearly came prepared to the job interview with all those details like the 800 weekly active users, and you, you wouldn't be, uh, surprised by maybe any question really. So good job. It was an awesome, um, op- opportunity to go into a head of a very demanding OpenAI, um, interviewer, and I don't know how other candidates would do, but definitely you'd be on my shortlist.

    2. AGAakash Gupta48:33

       Amazing. You heard it from him. He has been hiring manager at places like Microsoft AI. This would've passed. This is what a passing interview at OpenAI would look like. I have helped, just recently, I just helped a student sign a $656,000 OpenAI offer.So this is the process that I coached them through. They faced an interview with a very, very similar question. This is the type of question that OpenAI asks. This is the type of process that they were looking for. I want you guys to walk away with three key takeaways from what I did so that you can get inside your mind and operate like me. Let's start with takeaway number one. Takeaway number one, if you guys remember, I took a minute or two to create a unique framework. Do not come in and just use the same framework for every single question, but create a unique framework. Number two, do you guys remember when I added in these sections, what users are solving, prioritization and metrics based on what the interviewer was saying? Do you guys remember when I asked the interviewer, "How do you think this is going so far?" halfway through? You need to check in with your interviewer. You need to be very responsive to what they're saying. You can't-- A lot of people brush off what they're saying. They wouldn't have added in the new section to their framework. You need to go ahead and do that.

17. 50:03 – 52:02

    3 Takeaways You MUST Apply

    1. AGAakash Gupta50:03

       And then the third thing I did-

    2. BADr. Bart50:04

       And actually, I, I, maybe I'll add here because I think that some of the interviewers wouldn't be so open to give feedback mid-interview. But-

    3. AGAakash Gupta50:13

       Mm

    4. BADr. Bart50:13

       ... even if you don't get it, it's super important as a product manager to ask for feedback on whatever you do. So whether you get guidance or not, you'll can't fail by asking.

    5. AGAakash Gupta50:25

       Exactly. You can't fail by asking. It just shows that you are somebody who checks in. And in my practical experience, nine times out of 10, when you ask them, they do give you an honest answer. And so sometimes you just don't think that they would give you an honest answer. And the final bucket that I did, that 99% of you guys are not doing, because I do a lot of mock interviews, did you see how I related it back to my experience and my knowledge and my strengths? I consistently emphasized, "What is my unicorn candidate-market fit for OpenAI?" They don't have any PMs from gaming, so I'm gonna keep putting that in there into different points. I'm gonna talk about it. And what did I also say? "Oh, I've hit as big of users as you," because they want to see people who have done as big of scale, and so I justified that with my Google experience. And so make sure that you are dropping in elements of selling yourself and not just completing the case. That is how to get a job at OpenAI. If you guys want to learn more, if you guys want coaching from Bart and I three times a week for three months, we are taking a group of 30 elite PMs, application only, at landpmjob.com. We have already sold out half the cohort, so grab your seat, apply now. And if you like this content, follow, subscribe to this YouTube channel, and I'll see you in the next one.

    6. BADr. Bart51:59

       Thank you for having me. Awesome interview. See you next time.

Episode duration: 52:12