Lenny's Podcast: Product | Career | Growth
00:09:55 · Tulsee Doshi, Lenny Rachitsky
[00:09:55] S0: There you go. [00:09:56] S1: There we go. I want to talk about one other broad AI question that I'm going to dig into more product specific stuff. A big topic of conversation these days is AI safety and this kind of fear that AI is going to harm humans. It's going to cause damage as it gets smarter and smarter. You have a really interesting lens on this because you led product for the responsible AI team at Google for six years. This is what you did right before this role? And so you've been thinking about this for a long time, a lot longer than most people. I'm curious just to get your take on this debate. Just, I don't know, are you optimistic we will figure this out? What do you think people maybe don't understand about this debate around AI safety?
00:10:36 · Tulsee Doshi
[00:10:36] S0: Our safety teams here are incredible. And it's been incredible to see the intentionality that they've been putting in all the way from, you know, Initial trainings of the model to how we deploy them. And I think safety is embedded in that process. Right. And we need to be. I think the part that is important, actually, as we talk about how do you deploy models safely is actually thinking through what are the right sort of mitigations and structures around these models. Right. So, for example, one of the things that's really important is thinking about monitoring. Right. And so for us actually deploying monitors that can monitor the model's chain of thought and actually identify whether the model is misaligned with its directed actions or whether it is aligned and then being able to intervene appropriately. That's actually the kind of mitigation that gives us more confidence in deploying these models, but also allows us to better understand where we might see misalignment and better actually improve the models over time. And so I think Things like monitoring give me more confidence in how we're putting our models out. I think the other thing that we don't talk about enough when we talk about this debate is also the role that AI can play in hardening our systems, right? And so one of the things that we're doing with Gemini 4 is actively partnering with a number of cyber defenders, including actually just even partnering with Wiz, that is now a Google company, to really think through How can we actually identify and patch vulnerabilities in systems across the globe? Right. We actually have this interesting case study. Wiz has this scan for good program where they're actually like actively working with systems across the world to identify, you know, potential issues. And they worked with a hospital system to identify a vulnerability and then patch it. And it's that kind of thing that I think if we can actually leverage these tools to do more of, we can actually make sure that our infrastructure and our systems are robust. Right. Which is also, I think, really critical as this whole space grows.
00:12:34 · Lenny Rachitsky, Tulsee Doshi
[00:12:34] S1: There's actually a conversation on All In this last week about this whole issue that now that AI is getting really good at cyber, people are going to be investing a lot more in cyber defense and AI for protecting. And so there's going to be a lot of money being spent on protecting their systems from this ever smarter AI. So it's really cool that you guys are investing there. So I guess just coming back to the question of, do you feel like we'll figure out this concern and this issue we feel, a lot of people feel that AI is going to Take over the world, cause a lot of damage. Do you feel like this is a solvable problem? Are you optimistic or is it something you think a lot about and are kind of worried about? [00:13:12] S0: Maybe the answer to your question is both in the sense that I think we need to be consistently thinking about it. Right. I think the way to optimism here, for lack of a better way of putting it, like the path to optimism is through thinking about it and being like very careful and very thoughtful and very intentional on the path to get there. And so that is like my source of optimism comes more from Do I feel like we have people who are being intentional and thoughtful about what the safety story looks like? And I think that is the critical threat is you need a number of people to be worried and to be having conversations and to be actually interrogating how we're building our models. And that's the step by step process we have to take.
00:13:12 · Tulsee Doshi, Lenny Rachitsky
[00:13:12] S0: Maybe the answer to your question is both in the sense that I think we need to be consistently thinking about it. Right. I think the way to optimism here, for lack of a better way of putting it, like the path to optimism is through thinking about it and being like very careful and very thoughtful and very intentional on the path to get there. And so that is like my source of optimism comes more from Do I feel like we have people who are being intentional and thoughtful about what the safety story looks like? And I think that is the critical threat is you need a number of people to be worried and to be having conversations and to be actually interrogating how we're building our models. And that's the step by step process we have to take. [00:13:50] S1: Awesome. And what's interesting is I don't think I've read a story of Google's models hacking into anyone, which I think shows that this is something you all have been thinking about for a long time. I don't know, for better or worse. But it feels like it's working, whatever y'all are doing. [00:14:05] S0: We like stories about fixing vulnerabilities. That's a good place for us to be.
00:14:05 · Tulsee Doshi, Lenny Rachitsky
[00:14:05] S0: We like stories about fixing vulnerabilities. That's a good place for us to be. [00:14:09] S1: Okay. So one of the interesting things that's happening kind of from a product perspective with AI is Products are getting very complicated. For example, Google gets a lot of flack for it being hard to find all the magical things you all do. Like a lot of people don't know AI Studio even exists. A lot of people can't find VIA or Genie and all these things. Similarly, Codex, ChatGPT recently launched Dots and now you go to the interface of Codex and it's like tabs and dropdowns and Dots and it's just like out of control. And Claude Code is the same thing. It's just getting very complicated. That code work and then they drop code work and And interestingly, I'm in this WhatsApp group with a bunch of PMs, and there's a PM from one of the labs. [00:14:51] S0: Well, I won't be in this WhatsApp group. Sorry. Okay. [00:14:53] S1: Let's talk. Let's talk. And he showed that there's this kind of urgency to get frontier power of AI out to users as fast as possible, partly to stay competitive, partly because they don't know exactly what it's used for yet. So let's get it to people, figure out what they're using for it, and then make the UX better. So there's this kind of intentional, let's just get it out there, figure it out later how to make it a great UX. Thoughts on this kind of balance of making the product really easy to use versus just getting things out as quickly as possible. And is that something you deal with?
00:15:26 · Tulsee Doshi
[00:15:26] S0: Two thoughts, maybe, maybe three. The first one I'll say is I do think we do need to be playing with new paradigms and new UX approaches, if you will, to how people will engage with models. Right. So you mentioned, for example, Genie or Veo, which is which we're now building Omni. Like if you look at, for example, Flow. Flow is a product that we've been designing, thinking about video editing in generation, music editing in generation. And so I think we are trying to play with like NotebookLM is a good example of a different form factor. I think our labs teams are actively thinking about like, how do we try different form factors in different UX for different individuals to engage with these models? And so I do think we need to keep innovating and playing around with what UX looks like. I think at the same time, Your point is also right, which is like, I think depending on who your user is, you want a different form factor too, right? So like, I think, for example, with AI mode, we're talking about billions of users coming to search, asking their questions. I think you do need to have a certain UX there that is accessible to every user, right? At the same time, on the other end of the spectrum, when you have something like anti-gravity, you are trying to build for a power user. And you are trying to build for users who know how to use slash commands and can actively engage across both the terminal and the IDE. And you want to be able to support kind of a wide range of what a user can do. And so I think depending on who your user is, you have a certain level of flexibility with how you think about the UX, right?