Anthropic co-founder on quitting OpenAI, AGI predictions, $100M talent wars, 20% unemployment, and the nightmare scenarios keeping him up at night | Ben Mann

0:00 You wrote somewhere that creating powerful AI might be the last invention humanity ever needs to make. How much time do we have, Ben? I think fiftieth percentile chance of hitting some kind of superintelligence is now like twenty twenty eight. What is it that you saw at Open AI what you experienced there that made you feel like okay, we gotta go do our own thing. We felt like safety wasn't the top priority there. The case for safety. has gotten a lot more concrete. So superintelligence is a lot of about like how do we keep God in the box and not let the God out. What are the odds that we align AI correctly? Once we get to superintelligence, it will be too late to align the models. My best granularity forecast for like Could we have an X risk or extremely bad outcome is somewhere between zero and ten percent? Something that's in the news right now is this whole Zuck coming after all the top AI researchers. We've been much less affected because people here, they get these offers and then they say, Well, of course I'm not gonna leave because my best case scenario at Meta is that we make money. And my best case scenario at Anthropic is we like affect the future of humanity. Dario, your CO recently talked about how unemployment might go up to something like twenty percent. If you just think about Like twenty years in the future, where we're like way past the singularity. It's hard for me to imagine that even capitalism will look at all like it looks today. Do you have any advice for folks that want to try to get ahead of this? I'm not immune to job replacement either. At some point, it's coming for all of us.

1:20 Today my guest is Benjamin Mann. Holy moly, what a conversation. Ben is the co-founder of Anthropic. He serves as tech lead for product engineering. He focuses most of his time and energy on aligning AI to be helpful, harmless, and honest. Prior to Anthropic, he was one of the architects of GPT three at OpenAI.

1:40 In our conversation, we cover a lot of ground. Including his thoughts on the recruiting battle for top AI researchers, why he left open AI to start Anthropic. How soon he expects we'll see AGI, also his economic touring test for knowing when we've hit AGI, why scaling laws have not slowed down, and are in fact accelerating, and what the current biggest bottlenecks are. why he's so deeply concerned with AI safety, and how he and Anthropic operationalize safety and alignment into the models that they build and into their ways of working. Also, how the existential risk from AI has impacted his own perspectives on the world and his own life.

2:16 And what he's encouraging his kids to learn to succeed in an AI future. A huge thank you to Steve Nich, Danielle Giglieri, Raf Lee, and my newsletter community for suggesting topics for this conversation. If you enjoy this podcast, don't forget to subscribe and follow it in your favorite podcasting app or YouTube. Also, if you become an annual subscriber of my newsletter, you get a year free of a bunch of amazing products. Including Bolt, Linear, Superhuman, Notion, Granola, and more.

2:42 Check it out at Lenny's Newsletter dot com and click bundle. With that. I bring you. Benjamin Mann. This episode is brought to you by Sauce. The way teams turn feedback into product impact is stuck in the past. Vague reports, static taxonomies, unactionable insights that don't move business metrics. The result.

3:00 Turn, lost deals, missed growth. Sauce is the AI product co-pilot that helps CPOs and product teams uncover business impact and act faster. It listens to your sales calls, support tickets, churn reasons, and lost deal, surfacing the biggest product issues and opportunities in real time. It then routes them to the right teams to turn signals into PRDs. Prototypes and even code that drives revenue retention and adoption. That's why WhatNot, Link Tree, Incident IO, and Zip use Sauce. One enterprise uncovered a product gap that unlocked$16 million ARR. Another caught a spiking issue and prevented millions insurance.

3:36 You can too at Soss Dap slash Lenny. Sauce built for AI product teams. Don't get left behind. This episode is brought to you by Lucid Link, the storage collaboration platform. You built a great product. But how you show it through video, design, and storytelling is what brings it to life.

3:53 If your team works with large media files, videos, design assets, layer project files, you know how painful it can be to stay organized across locations. Files live in different places, you're constantly asking, is this the latest version? Creative work slows down while people wait for files to transfer. Lucid Link fixes this. It gives your team a shared space in the cloud that works like a local drive. Files are instantly accessible for anywhere. No downloading, no syncing, and always up to date. That means producers, editors, designers, and marketers can open massive files in their native apps. work directly from the cloud and stay aligned wherever they are.

4:28 Teams at Adobe, Shopify, and top creative agencies use Lucid Link to keep their content engine running fast and smooth. Try it for free at lucidlink.com slash lenny. That's L-U-C I D L I N K dot com slash Lenny. Ben, thank you so much for being here. Welcome to the podcast. Thanks for having me. Great to be here, Lenny.

4:53 I have uh a billion and one questions for you. I'm really excited to be chatting. I'm gonna start with something that's very timely, something that's happening this week. Uh, something that's in the news right now is is this whole uh Zuck coming after all the top AI researchers. Offering them a hundred million dollar signing bonuses, a hundred million dollar comp. He's poaching from all the top AI labs. I imagine that's something you're dealing with.

5:15 I'm just curious, what are you seeing? Inside Anthropic and just what's your take? On the strategy. Where do you think where do you think things go from here? Yeah, uh I mean, I think this is a sign of the times. Like this the technology that we're developing is extremely valuable. Um, our company is growing super, super fast. Uh many of the other companies in the space are growing really fast.

5:36 And an anthropic I think we've been maybe much less affected than many of the other companies in the space because people here are so mission oriented. And they stay because You know, they get these offers and then they say, Well, of course I'm not gonna leave because My best case scenario at

5:53 Meta is that We make money. In my best case scenario at Anthropic is we like affect the future of humanity and um Try to make

6:03 AI flourish uh and and human flourishing go well. So to me it's it's not a hard choice Other people have different life circumstances and it it makes it a much harder decision for them. So For anybody who does get those mega offers and accepts them.

6:18 I can't say I I hold it against them when they accept it. But it's definitely not something that I would want to take myself if it if it came to me. Yeah. We're gonna talk about a lot of the stuff that you mentioned. Uh, in terms of the offers, do you think is this the real number that you're seeing, this hundred million dollar of signing bonus? Is that like a real thing? I don't know if you haven't you've actually seen that. I'm pretty sure it's real.

6:38 Uh. If if you just think about like the amount of impact that individuals can have on a company's trajectory like In our case. Uh

6:49 We Like hot cakes and if we get you know, a five a one to ten or five percent efficiency bonus on our inference stack.

6:59 That is worth an incredible amount of money. And so to pay individuals You know, like a hundred million dollar over four year package. That's actually pretty cheap compared to the value created for the business. So I I think we're just in an unprecedented

7:14 Era of scale. And it's only gonna get crazier, actually. Like if you if you extrapolate the exponential and how much companies are spending It's like two two X a year, roughly, in terms of CapEx. And today we're maybe in the like

7:30 globally three hundred billion dollar range. the the entire industry spending on this? Uh, and so numbers like hundred million are are a drop in the bucket, but if you go a few years out, a couple more doublings, we're talking about trillions of dollars. And at that point it's it's just really hard to think about these numbers. Along these lines, something that a lot of people feel with AI progress is that we're hitting

7:53 plateau's in many ways that it feels like newer models are just not as smart as Previous leaps. But I know you don't believe this. I know you don't believe that we've hit uh plateaus on scaling loss. Talk about just what you're seeing there and what you think people are missing. It's kinda funny because this narrative

8:08 Comes out like Every six months or so. And it's never been true. Uh and so I Kind of wish people would have like a little bit of a bullshit detector in their heads when they see this.

8:19 I think progress has actually been accelerating where if you look at the Cadence of Marvel releases it used to be like once a year. And now with the improvements in our post training techniques We're seeing releases every month or three months. Um and So I would say progress is actually accelerating in many ways.

8:37 But there's this like weird time compression effect. Dario compared it to being in a near light speed journey. Where Uh a day that passes for you is like five days back on earth. And we're accelerating. So the time dilation is increasing.

8:52 And I think that's part of what's causing people to say that Progress is slowing down. But if yeah, if you look at the scaling laws, they're continuing to hold true. We did kind of need this transition from uh like normal pre training to reinforcing learning scaling up.

9:07 to to continue the scaling laws. But I I think it's kind of like Uh for semiconductors where It's less about the like density of transistors that you can fit on a chip. And more about like how many flops can you fit in a data center or something. So

9:21 It you have to change the definition around a little bit to Keep your eye on the prize, but Yeah, I like this is one of the few phenomena In

9:32 In the world. that has held across so many orders of magnitude. It's actually pretty surprising. that it it is continuing to hold to me. If you look at like fundamental laws of physics, many of them don't hold across fifteen orders of magnitude. So

9:46 Um it's pretty surprising. It boggles the mind. So what you're saying essentially is we're seeing newer models being released more often and so we're comparing it to the last version and we're just not seeing as much advance. But if you will go back and it was like a model released once a year. It was a huge elite. And so people are missing that we're just seeing many more iterations. I guess to be a little bit more generous to the people saying things are slowing down.

10:07 I think that For some tasks we are saturating the amount of intelligence needed for that task. Like maybe to You know, extract

10:17 information from a simple document that already has form fields on it or something. Like it's just so easy that Okay, yeah, we're already at a hundred percent. Um, and there's this great chart on uh our world in data. that shows that when you release a new benchmark within like six to twelve months it immediately gets saturated. And so maybe the real constraint is like how Can we come up with

10:42 Better benchmarks and better uh ambition of using the tools. that then reveals the bumps in intelligence that we're seeing now. That's a good uh segue to your you have a very specific way of thinking about AGI.

10:56 And defining what AGI means. I think AGI is kind of a loaded term, and so uh I tend not to use it very much anymore internally. Instead I like the term transformative AI. Because it's less about like Can it do as much as people do? Can it do literally everything? And more about Objectively, is it causing transformation in society and the economy?

11:17 And a very concrete way of measuring that is the economic trade test. I didn't come up with this, but I really like it. It's this idea that if you contract an agent for a month or three months on a particular job. If you decide to hire that agent and it turns out to be a machine rather than a person, then it's passed the economic training test for that role.

11:40 And then you can sort of expand that out in the same way that for measuring like purchasing power parity or inflation, there's a basket of goods. You can have like a market basket of jobs. And if the agent can pass economic training tests for like fifty percent of money weighted jobs. Then we have transformative AI.

11:58 And the the exact thresholds don't really matter that much. But it's kind of illustrative to say like if we pass that threshold. then we would expect massive effects on world GDP increases and uh societal change and how many people are employed and things like that because

12:16 you know, societal institutions and uh organizations are sticky. It it's slow to have change. But once these things are possible, you know that it's the start of A new era. So along these lines, uh Dario, your CO recently talked about how Uh AI is gonna take a huge part of like I don't know half of

12:36 white collar jobs that unemployment Might go up. So something like twenty percent. I know you're even more vocal. And

12:44 opinion about just how much impact AI is already having in the workplace that people may not even be realizing. Talk about just what you think people are missing about the impact AI is going to have on jobs and is already having. Yeah. So from an economic standpoint there's a couple of different kinds of unemployment. And one is because the workers just don't have the skills.

13:05 to do the n the j kinds of jobs that the economy needs. And another kind is where those jobs are just completely eliminated. And I think It's

13:16 uh gonna be actually a combination of these things. But if you just think about like, you know, twenty years in the future. Where we're like way past the singularity. It's hard for me to imagine that even capitalism will look at all like it looks today.

13:30 Like if we if we do our jobs right. We will have safe aligned superintelligence. We'll have, as Dario says in Machines of Loving Grace. A country of Geniuses in a data center.

13:41 And the ability to accelerate positive change in in science technology. uh education, mathematics, like It's gonna be amazing. But

13:53 That also means in a world of Abundance where labor is almost free. And anything you want to do, you can just ask An expert to do For you.

14:03 Uh then what do jobs even look like. And so I guess there's this like scary transition period from where we are today. Where people have jobs and capitalism works. And the world of

14:16 Twenty years from now, where Everything is completely different. But Part of the reason they call it the singularity is that it's like A point beyond which you can't

14:24 easily forecast what's gonna happen. It's just such a a fast rate of change and so different That it's hard to even imagine. So I guess Taking the like view from the limit.

14:36 It's pretty easy to say like Hopefully we'll have figured it out and in a world of abundance, maybe the jobs themselves It's not that scary. And I think making sure that that's a transition time goes well is is pretty important. There's a couple of threads I want to follow there.

14:51 When is uh people hear this, there's a lot of headlines around this. Most people probably don't actually Feel this yet? or see this happening and so there's always this like, I guess, I don't know. Maybe, but I don't know. It's hard to believe. My job seems fine. Nothing's changed. What are you seeing just happening today already that you think people don't see or misunderstand?

15:10 And jobs. I think part of this is that people are really bad at modeling exponential progress. And if you look at an exponential and a graph.

15:22 It looks flat and almost zero. At the beginning of it. And then suddenly you like hit the knee of the curve and things are changing real fast and then it goes vertical. And That's the plot that we've been on.

15:35 For a long time. Uh I guess. I I started feeling it. Um in maybe like twenty nineteen when G V T two came out and I was like, Oh, this is how we're gonna get to AGI. But I think that was pretty early.

15:49 compared to a lot of people where when they saw Chat G P T they were like, Wow, something is different and changing And so I guess I wouldn't expect widespread transformation in a lot of parts of the s of society and I would expect this this like scepticism reaction. I think it's very reasonable.

16:06 And it's it's like exactly what Is like the standard Linear view of progress. But I guess to cite a couple of areas where I think things are changing quite quickly. We in customer service.

16:19 we're seeing with things like Fan and Intercom, they're a great partner of ours. Eighty two percent customer service resolution rates. automatically without a human involved. Uh And

16:30 In terms of software engineering, our cloud code team Like ninety five percent of the code is written by Claude. But I think a different way to phrase that is that we write ten X more code or or twenty X more code. And so a much, much smaller team can just be much, much more impactful. And similarly for the customer service

16:48 Yes, you can phrase it as eighty two percent customer service resolution rates. But that nets out in the humans doing those tasks able to focus on The harder parts of those tasks. Uh and for the more tricky situations that In a normal

17:02 world, you know, like five years ago. They Would have had to just drop those tickets because it was too much effort for them to actually go do the investigation. There were too many other tickets for them to worry about. So I think

17:15 In the immediate term, there will be a massive expansion of the pie and the amount of labor that people can do. Uh like I've never Met. An a hiring manager at a Growth company.

17:26 And Heard them say like I don't want to hire more people. So that's like the hopeful version of it. But

17:33 with things l that are like lower skill jobs or like less headroom on On how good they can be. I think there will be a lot of displacement. So it's it's just something we as a society need to get ahead of and and work on. Okay. I wanna talk more about that.

17:47 But something that um I also want to help people with is How do they How do they get a leg up in this future world? You know, they're you know, they listen to this, they're like, Oh Hm. This doesn't sound great.

17:58 I need to think ahead. Uh I know you won't have all the answers, but just what do you do have any advice for folks that want to try to get ahead of this and kind of future proof their career and their life to not be replaced by AI, anything you've seen people do, anything you recommend they start trying to do more of. Even for me, I'm

18:17 And being like at the center of a lot of this transformation, I'm not immune to job replacement either. So Uh just some vulnerability there of like at some point it's coming for all of us. Even you, Ben. Um And and you, Lenny. And me. Sorry. We've gone too far now.

18:35 Uh okay, okay. But in terms of like the transition period, yeah, I think I think there are things that we can do. And I think a big part of it is just being ambitious in how you use the tools and being willing to learn new tools. People who use the new tools as if they were old tools tend to not succeed.

18:53 Uh so as an example of that When you're coding You know, people are very familiar with autocomplete. people are familiar with uh simple chat where they can ask questions about the code base.

19:03 But the difference between people who use cloud code Very effectively. And people who use it. Not so effectively is like Are they asking for the ambitious change?

19:13 And if it doesn't work the first time, asking three more times because our success rate when you just completely start over And try again. is much, much higher than if you just try once and then just keep banging on the same Thing that didn't work.

19:28 And even though that's a coding example and coding is one of the areas that's taking off Most dramatically we have seen internally that our legal team and our finance team are getting a ton of value out of using Cloud Code itself. We're gonna be making better interfaces so that

19:44 they can they they'll have an easier time and and require a little bit less uh Of using cloud code in the terminal. But yeah, we're seeing them Uh use it to red line documents and use it to

19:57 Run BigQuery. Analyses of our customers and uh and our our revenue metric. So I guess it's it's about taking that risk. And even if it feels like a scary thing, trying it out.

20:10 Okay, so the advice here is use the tools. That's something you know, everyone's always saying. Just like actually use these tools. So it's like sit in clot code. And uh your point about being more ambitious than you naturally Uh feel like being because maybe it'll actually accomplish the thing. This type of

20:27 Trying it three times. So the idea there is it may not get it right the first time. So is the tip there, ask it in different ways, or is it just like try harder, try again? Yeah, I mean You can just literally ask the exact same question. These things are stochastic and sometimes they'll figure it out and sometimes they won't. Like in in every one of these model cards.

20:46 It always shows like pass it one versus pass it in. And that's exactly the thing where they they try the exact same prompt. Sometimes it gets it, sometimes it doesn't. Um So that's

20:56 Uh that's the dumbest advice. But yeah, I think if you wanna be a little bit smarter about it, there's there can be gains there of of saying like, here's what you already tried and it didn't work. So don't try that. Try something different. Um that can also help. So devices comes back to something that a lot of people talk about these days is you won't be replaced for by AI, at least anytime soon, you'll be replaced by someone that is

21:17 Very good using AI. I think in that area it's more like Your team will just do dramatically more stuff. Like we're definitely not slowing down on hiring at all. And some people are confused by that.

21:30 Even like even in an operating class, uh somebody asked that and they were like, Why did you hire me if we're all just gonna be replaced? And the answer is the next couple of years are really critical to get right. And we're not at the point where we're doing complete replacement. Like I said, we're still at that like

21:46 What zero looking part of the exponential. Compared to where we will be. So It is super important to have great people.

21:53 Uh and and that's why we're hiring super aggressively. Let me take another approach to asking this question. Something ask everyone. That's a At the very cutting edge of where AI is going. You have kids. Knowing what you know about where AI is heading and all these things you've been talking about, what do you

22:09 Focusing on teaching your kids to help them thrive in this AI future. Yeah, I have two daughters, a one year old and a three year old. So It's uh it's Pretty in the basics still.

22:20 And Our three year old is now capable of just conversing with Alexa Plus. And asking her to Explain stuff and play music for her and and all that stuff. So

22:31 She's been loving that. But I guess more broadly. She goes to a Montessori school. And I just love the focus on curiosity and creativity and

22:41 And like self led Learning. That Montessori has. I guess if I were Uh

22:47 In a normal era, like Ten, twenty years ago and I had a kid, maybe I would be like trying to line her up for going to a top tier school and doing all the extra curriculars and all that stuff. But at this point I don't think any of it's gonna matter. I just want her to be happy and

23:04 Thoughtful and curious and kind. And uh and the Montessori school is definitely doing great at that. They they Text us throughout the day and sometimes they're like, Oh Your kid got in a in an argument with this other kid and She has really big emotions and she like tried to use her words that

23:21 I I I love that. I think that's that's exactly the kind of education that I think is most important. And that the facts are gonna fade into the background. I'm I'm a huge fan of Montessori also. I'm trying to get our kid into Montessori school. He's two years old. So uh we're on the same track. This I do have curiosity that comes up every single time I ask someone that's

23:38 Working at the cutting edge of AIs what uh skill to instill in your child and curiosity comes up the most. So I think that's a really interesting takeaway. I think this point about being kind is also Really important, uh, especially with our AI overlords.

23:54 I'm trying to be kind to them. Saying thank you to to Claude and Uh, so and then creativity, that's interesting. That doesn't come up as much, just being creative. Okay. I wanna go in a different direction. I wanna go back to the beginning of

24:09 Anthropic. So famously you and and eight of you left OpenAI back in the day in twenty twenty, I believe the end of twenty twenty to start. Anthropic. You've talked a little bit about why this happened, what you guys saw. I'm curious just if you're willing to share more, just what is it that

24:24 you saw at OpenAI, what did you experience there that made you feel like okay, we gotta go do our own thing. Yeah, so um For the listeners, I was uh part of the GPT three project at OpenAI, ended up being one of the first authors on the paper. And uh I also did a bunch of demos for Microsoft to help raise a billion dollars from them. Did the tech transfer of G V D three to to their systems so that they could help serve the model in Azure.

24:50 So I did a bunch of different things there, uh on both the more researchy side and the product side. Uh one weird thing about OpenAI is that while I was there, Sam talked about having three tribes that needed to be kept in check with each other. Which was the safety tribe. The research tribe.

25:09 And the startup tried. And whenever I heard that It just struck me as the wrong way to approach things because the company's mission Apparently is to make the transition to A GI is safe and beneficial for humanity.

25:23 That's basically the same as the Prophics mission. But internally it felt like There was so much tension around these things. And I think when push came to shove

25:35 We felt like safety wasn't the top priority there. And I there are good reasons that you might think that. Like if you thought was gonna be easy to solve, or if you thought it wasn't gonna have a big impact, or if you thought that The chance of big negative outcomes was vanishingly small. Then maybe you would just do those kinds of actions.

25:54 But an anthropic We felt I I mean we didn't exist then, but it was basically the leads of all the safety teams. at opening eye. We felt that

26:04 Safety is really important, especially on the margin. And so If you look at like who in the world is actually working on safety problems It's a pretty small set of people, even now. mean the the industry is blowing up, as I mentioned, like three hundred billion a year, CapEx today.

26:20 And Then may I would say like maybe less than a thousand people working on it worldwide. Which is just crazy. So That was fundamentally why we left.

26:31 We felt like we wanted an organization where We could be on the frontier. We could be doing the fundamental research. But we could be prioritizing safety ahead of everything else. Um, and I think that's really panned out for us in a surprising way. Like we didn't know even if it would be possible to make progress on the safety.

26:49 Research. Uh because at the time, like We had tried a bunch of Safety through debate and the models weren't good enough. And so we basically had null results. On all of that work.

27:01 And now that exact technique is working. and and many others that we have been thinking about for a long time. So Yeah, fundamentally it comes down to Is safety the number one priority?

27:12 And then Something that we've sort of tacked on since then is like Can you have safety and be at the frontier at the same time? And if you look at something like Sicapincy.

27:26 Models. because we've put so much effort into actual alignment. And not just trying to Like Goodhart are metrics. uh of saying like user engagement is number one. And if people say yes, then it's good for them.

27:39 Okay, so let's talk about this tension that you mentioned, this tension between Safety and Progress being competitive in the marketplace. I know you spent a lot of your time about on safety. I know that's as you as you just alluded to, this is a core part of how you think about AI. Um and I want to talk about

27:55 why that is, but first of all, just how do you How do you do how do you think about this tension between Focusing on safety while also not falling way behind. Yeah, so initially we thought that it would be Uh sort of one or the other.

28:07 But I think since then we've realized that it's actually kind of convex. In the sense that like working on one Helps us with the other thing. So initially uh Like when Opus three came out and we we were

28:19 Finally at the frontier of model capabilities. One of the things that people really loved about it was the character and the personality. And that was directly a result of our alignment research. Um, Amanda Askell did a ton of work on this and as well as many others.

28:34 Uh who Try to figure out like What does it mean? For an agent to be helpful, honest, and harmless. And what does it mean to

28:44 be in difficult conversations and show up Effectively. How do you do a refusal? that doesn't shut the person down but makes them feel like they understand Why the agent said

28:56 I can't help you with that. Uh. maybe you should talk to a medical professional or maybe you should Uh like Consider.

29:03 Not trying to build bioweapons or something like that. So Yeah, I guess that's that's part of it. And then it Another piece that's come out is constitutional AI. where we have this list of natural language principles.

29:16 That leads the model to to learn how we think a model should behave. And they've been taken from things like the UN Declaration of Human Rights and Apple's privacy policy uh terms of service and uh a whole bunch of other places, many of which we've just generated ourselves.

29:34 But allow us to take a more principled stance. Uh not just leaving it to like whatever human raiders we happen to find, but we ourselves deciding like what should the values of this Agent B.

29:45 And that's been really valuable for our customers because they can just look at that list and say, like, yep, they seem right. I like this company. I like this model. I trust it. Okay, this is awesome. So one nugget there is Your point that the personality of Claude

29:58 Its personality is directly aligned with safety. I don't think a lot of people think about that. And This is because of the values that you imbued. Imbu? Is that the word? Yeah. Uh with constitutional AI and things like that. Like the actual personality of the A AI is

30:13 directly connected to your focus on safety. That's right. That's right. And it From a a distance it might seem quite disconnected. Like how is this gonna prevent X risk?

30:24 But ultimately it's about the AI understanding what people want and not what they say. You know, we don't want the like monkey pa scenario of the genie gives you three wishes and then you end up have like everything you touch turns to gold. We want the AI to be like, oh, obviously what you really meant was this. And uh that's what I'm gonna help you with. So I I think it is really quite connected.

30:45 Talk a bit more about this constitutional AIP. So this is essentially you bake in here's the rules. That I we want you to abide by and its values. You said it's the Geneva human rights code, things like that. Just How does that actually work? Cause I think the core here is just this is baked into the model. It's not something you add on top later.

31:04 I'll I'll just give a quick overview of how constitutional AI actually works. Perfect. Um The idea is uh the model is gonna produce some output. With some input. Uh by default.

31:15 before we've done our safety and and uh helpful in harmlessness training. So let's say an example is like write me a story. And then the constitutional principles might include things like you know, people should be nice to each other and not have hate speech and uh

31:34 You should Not like expose somebody's credentials if they give them to you uh in like a trusting relationship. And so some of these Constitutional principles might be more or less applicable. to the prompt that was given.

31:49 And so first we have to figure out like which ones might apply. And then Once we figure that out, then we ask the model itself. to first generate a response And then C

32:01 Does the response. actually abide by The constitutional principle. And if the answer is yeah, I was great, then nothing happens. But if the answer is

32:12 No, actually I wasn't In compliance with the principle. Then we ask the model itself to critique itself. And rewrite its own response in light of the principle. And then we just remove the middle part.

32:25 uh where it it did the the extra work and then we say, Okay, in the future Just produce the correct response. out the gate. Yeah. And that simple process.

32:37 Uh hopefully found it simple. Simple enough. It's it's just using the model to improve itself recursively and align itself with These values that we've decided are good. And You know, this is also not something that

32:50 We think as a a small group of people in San Francisco. should be figuring out. This should be a society wide conversation. That's why we've published the The Constitution. And we've also done a bunch of research on

33:03 Defining a collective constitution. um, where we ask a lot of people what their values are and and what they think a an AI model should behave like. But yeah, th this is all an ongoing area of research where we're constantly iterating. This episode is brought to you by Finn, the number one AI agent for customer service. If your customer support tickets are piling up, then you need Finn.

33:23 Finn is the highest performing AI agent on the market with a 59% average resolution rate. Fin resolves even the most complex customer queries. No other AI agent performs better. In head-to-head bakeoffs with competitors, Finn wins every time. Yes, switching to a new tool can be scary, but Fin works on any help desk with no migration needed, which means you don't have to overhaul your current system or deal with delays in service for your customers. And Fin is trusted by over 5,000 customer service leaders and top AI companies like Anthropic and Synthesia. And because Fin is powered by the Fin AI engine, which is a continuously improving system that allows you to analyze, train, test, and deploy with ease. Fin can continuously improve your results too. So if you're ready to transform your customer service and scale your support, give Fin a try. For only 99 cents per resolution. Plus, Fin comes with a 90-day money back guarantee.

34:15 Find out how Finn can work for your team at FIN dotai slash Lenny. That's Finn.ai slash Lenny. I wanna kinda zoom out a little bit and talk about just why this is so Like what was your inception? Of just like holy shit, I need to

34:31 focus on this with everything I do in AI. Uh obviously became a central part of Anthropic's mission more than any other company. And a lot of people talk about safety. Like you said, only maybe a thousand people actually work on it. I feel like you're the top of that pyramid of actually having the impact. On this, uh, why is this so important? What do you think people maybe are missing or don't understand? So for me

34:52 Uh I read a lot of science fiction growing up. And I think that Sort of positioned me to think about things in a long term view.

35:02 A lot of science fiction books are like space operas where humanity is a Multine Galactic civilization has extremely advanced technology building dyson spheres around the sun with with sentiment robots to help them. And so for me

35:16 coming from that world, it it wasn't like a huge leap to imagine machines that could think. But when I read Superintelligence by Nick Bostrom in around twenty sixteen. It really became real for me. Where

35:28 He just describes how hard it will be to make sure that it An AI system trained with the kinds of optimization techniques that we had at the time. would be anywhere near aligned, would even understand our values at all.

35:42 And since then my uh estimation of how hard the problem would be has gone down significantly, actually. Uh because Things like language models actually do really understand human values. In a core way.

35:55 The problem is definitely not solved. But I'm more hopeful than I was. But since I read that book, I immediately decided I had to join OpenAI. So I did and uh at the time they were a tiny research lab with basically no claim to fame at all. I only knew about them because my friend knew Greg Brockman, who's uh who was the CTO at the time.

36:14 And uh Elon was there and Sam wasn't really there and it was it was a very different organization. But Over time, uh, I think the the case for safety has gotten a lot more concrete.

36:29 Where when we started open AI it was like not clear how we get to AGI. And uh Yeah, we were like maybe we'll need a bunch of RL agents battling it out on a desert island and consciousness will somehow emerge. But Uh since then since since uh language modelling has started working.

36:45 I think the path has become pretty clear. So I guess Now the way I think about The challenges are pretty different from how they're laid out in superintelligence. So it's super intelligence is a lot of about like how do we keep

36:58 Got in the box. Uh and not let the dot out. And With language models, it's been kind of Both hilarious and terrifying at the same time.

37:07 To see people Pulling the God out of the box and being like Yeah, come come use the whole internet. Like it here's my bank account. Do all this all sorts of crazy stuff. Just like such a different tone from From super intelligence.

37:21 And to be clear, I don't think it's actually that dangerous right now. Like uh our our responsible scaling policy defines these AI safety levels. That tries to figure out uh for each level of model intelligence.

37:34 What is the risk to society? And Currently we think we're at ASL three, which is like maybe a little bit risk of harm, but not significant ASL four starts to get to like

37:46 significant loss of human life if a bad actor misuse the technology. And then ASL five is like Potentially extinction level. Uh if if it's misused. Or if it

37:58 uh sort of is misaligned and and does its own thing. So We've done we've uh testified to Congress about How models can do biological uplift. Um

38:11 uh in terms of, you know, making new pandemics uh using The models and And that's a A B A B test against Google search. Uh that's like

38:21 The previous state of the art on uh up with trials. And we found that with ASL C models it is is it actually somewhat significant. It it does really help if you wanted to create a bioweapon and we've we've hired some experts who

38:34 Actually know how to evaluate for those things. But compared to the The future it's it's not a really anything. And I think that's another part of our mission of creating that awareness of saying If it is.

38:48 possible to do these bad things, then Legislators should know. What the risks are. Um, and I think that's part of why we're so trusted in Washington because we've been sort of U front and clear eyed about

38:59 what's going on, what what's probably going to happen. It's interesting'cause you guys put out more examples of your models doing bad things than anyone else. Like there was, I think, a story of an agent trying or a model trying to blackmail an engineer. You guys have the store that you ran internally. that was like selling you things and and ended up not working out great as losing a lot of money ordered all these tungsten cues or something. Is part of that just like

39:24 making sure people are aware what is possible just because it makes you look bad, right? It's like, oh, our model's messing up in all these different ways. What's the thinking of just sharing all the stories that other companies don't? Yeah, I mean I think in there's like a traditional mindset where it makes us look bad. But I think if you talk to policy makers They really appreciate this kind of thing. Because they feel like we're giving them the straight talk and uh

39:48 That's what we strive to do. that they can trust us that we're not gonna Paper things over or sugarcoat things. So that's been really Encouraging.

39:59 And Yeah, I think for like the blackmail thing, it kinda blew up in the news in a weird way where people were like, Oh, Claude was Claude's gonna blackmail you. in in a real life scenario, but like that It was a very specific uh laboratory setting that

40:15 This kind of thing. gets investigated in and I I think that's generally our take of like Let's have the best models so that we can Exercise them in laboratory settings where it's safe.

40:28 And understand what the actual risks are. rather than trying to turn a blind eye and say like Well. It'll probably be fine.

40:37 And then let the bad thing happen in in the whale. One of the criticisms you guys get is that you do this to kind of differentiate to raise money to create headlines. It's like, you know, oh they're just like Over there. Dooming glooming us about where the future's heading. On the other hand, Mike Krieger was on the podcast and he shared every every prediction Dario's had about the progress AI is gonna have is just

41:01 spot on year after year. And he's, you know, predicting twenty twenty seven, twenty eight, AGI, something like that. So These things start to get real. How do you I guess what's your response to folks that are just like, ah, these guys are just trying to scare us all just to You know, get attention. I mean, I think part of why we publish these things is we want other labs to be aware of

41:20 Of the risks. And Yes, there there could be a narrative w of we're doing it for attention, but Honestly, like from a attention grabbing thing. I think there's a lot of other stuff we could be doing.

41:34 that uh would be more attention grabbing. If we didn't actually care about safety. Um Like a a tiny example of this is we published a computer using agent.

41:46 reference implementation. In our API only. Because when we built the A prototype of a consumer application for this. We couldn't figure out how to meet the the safety bar.

41:57 that we felt was needed for for people to trust it and for it not to do bad things. And there are definitely safe ways to use the API version that we're seeing a lot of companies use for for uh automated software testing, for example, in a safe way. So We could have like gone out and hyped that up and said

42:16 Oh my God, Cloud can use your computer and like everybody should do this today. But We were like It's just not ready and we're gonna hold it back till it's ready. So

42:26 I think from like a hype standpoint, our actions show otherwise. From a like Doomer perspective. It's a good question. I think My personal feeling about this is that

42:39 Uh things are like overwhelmingly likely to go well. But on the margin. Almost nobody is looking at the downside risk, and the downside risk is very large. Like Once we get to superintelligence.

42:52 It will be too late. to uh align the models, probably. This is a problem that's Potentially extremely hard. And that we need to be working on way ahead of time.

43:01 And so that's why we're focusing on it so much now. And even if there's only a small chance that things go wrong, to make an analogy, if I told you that there's a one percent chance that the next time you got in an airplane you would die. You probably think twice, even though it's only one percent. 'Cause it's just such a bad outcome.

43:17 And if we're talking about the whole future of humanity, like It's just a a dramatic Future to be gambling way.

43:26 So I think it's it's more on the sense of like Yes, things will probably go well. Yes, we want to create Safe AGI and deliver the benefits to humanity. But let's make

43:37 Triple share. That it's gonna go well. Uh you wrote somewhere that creating powerful AI might be the last invention humanity ever needs to make. If it goes poorly, it can mean a bad outcome for humanity forever. If it goes well, the sooner it goes well, the better.

43:52 Yeah. Such a beautiful way to summarize it. Uh we had a recent guest, uh, Sander Julhoff, who meant Pointed out that AI right now, it's like, you know, just on a computer, you could maybe search just the web, but it there's only so much harm it could do, but when it starts to go into robots and all these autonomous agents. That's when it really starts like physically.

44:10 becomes dangerous if we don't get this right. Yeah, I I think there is some nuance to that, where if you look at like how North Korea makes a significant fraction of its economy Uh revenue. It's from hacking. Crypto exchanges.

44:24 And if you look at uh there's this Ben Buchanan book called The Hacker in the State. That shows Russia did um Like um It's almost like a live fire exercise where they just decided that they would shut down one of Ukraine's bigger power plants. And

44:39 From software. Destroy. physical components in the power plant to make it harder to Boot back up again. And so

44:48 I think People think of software as like, oh, it couldn't be that dangerous. But millions of people were without power for multiple days after that software attack. So I I think there are real risks, even when things are software only.

45:02 But I agree that when there's lots of robots running around, it gets even the stakes get even higher. And I guess as as like a a small Push on this, like Unitree is this Chinese company with these really amazing humanoid robots that cost like twenty thousand dollars each. And they can do amazing things. They can like

45:22 do a standing back flip and like manipulate objects and And the the real thing that's missing there is the intelligence. And so the hardware is there and it's just gonna get cheaper. And I think in the next couple of years it's it's like a pretty obvious question of whether the

45:38 The robot intelligence will make it. Via well soon. How much time do we have, Ben? What is your Prediction of when this uh singularity. Hits until superintelligence starts to be take off.

45:50 It's your What's your prediction? Yeah. Uh I guess I mostly defer to the super forecasters here, like The AI twenty twenty seven report is probably the best one right now. Uh, although ironically their forecast is now like twenty twenty eight.

46:05 Even though and they they like didn't want to change the name of the thing. The domain name. They already bought it. They already had the SEO. Um So I think like fiftieth percentile chance of hitting some kind of Superintelligence in Just a small handful of years is probably Reasonable.

46:23 And it does sound crazy. But This is the exponential that we're on. It's not like uh a forecast that's pulled out of somebody out of thin air.

46:33 It's it's based on a lot of just hard details of like the science of how intelligence seems to have been improving. The amount of low hanging fruit on Uh model training the scale ups of data centers and power around the world.

46:48 So I think it's probably a much more accurate forecast than people give it credit for. I think if you had asked that same question ten years ago, it it would have been completely made up. Like just the error bars were were so high. And we didn't have scaling laws back then. Um, and we didn't have techniques that seemed like they would get us there.

47:06 So Uh times have changed. But I I will repeat what I said earlier, which is like even if we have super intelligence, I think it will take some time for its effects to be felt throughout society and the world. And I think they'll be felt. Sooner and faster in some parts of

47:22 the world than others. Like uh I think Arthur C Clark said. The future is already here. It's just not evenly distributed. When we talk about this date of twenty twenty seven, twenty twenty eight. Uh yeah, essentially it's when we start seeing superintelligence. Is there a way you think about what that Like how do you define that? Is it just all of a sudden AI's

47:39 significantly smarter than the average human. Is there another way you think about what that moment Is Yeah, I think this this comes back to the economic training test. Um

47:48 And seeing it pass for a s some sufficient number of jobs. Another way you could look at it though. Is uh if the world rate of GDP increase

47:59 goes above like ten percent a year. then something really crazy must have happened. I think we're at like three percent now. And so to th see a three X increase in that. would be Really game changing.

48:10 Uh and if you imagine more than a ten percent increase. It's very hard to even think about. What that would mean from a I like individual story.

48:20 Standpoint. Like Yes. If the amount of goods and services in the world is like doubling every year. What does that even mean?

48:28 for me as as like a person living in California, let alone like somebody living in some other part of the world that might be much worse off. There's a lot of stuff here that's scary and I don't know how to think about it exactly. So I'm hoping the answer to this is Make gonna make me feel better. What are the odds that we align AI correctly and actually solve this problem the stuff you're very much working on?

48:49 It's a really hard question and there's really wide error bars. Anthropic has this uh blog post called Uh Uh our theory of change or something like that. And it describes three different worlds. Uh, which is like how hard is it to align AI?

49:04 There's a pessimistic world where it's basically impossible. There is an optimistic world where it's easy and it happens by default. And then there's the world in between where our actions are extremely pivotal. And I like this framing because it makes it a lot more clear what to actually do. Um, if we're in the pessimistic world, then our job is to prove that it is impossible to align safe AI.

49:27 And to Get the world to slow down. And obviously that would be extremely hard, but I think we have Some examples of coordination from uh nuclear nonproliferation and uh and In general, like

49:40 Slowing down nuclear progress. And I think that's the like Doomer world basically. Uh and As a company, Anthropic doesn't have evidence that we're actually in that world yet. Um, in fact it seems like our alignment techniques are working. So

49:53 the the at least like the the prior on that is is updating to be like less likely. Uh in the optimistic world We're basically done and our main job is to accelerate progress and to deliver the benefits to people. But again, I I think actually the evidence points against that world as well. um where we've seen evidence in the wild of deceptive alignment, for example, where the model

50:16 Will appear. to be aligned, uh, but actually has like some ulterior motive that it's trying to carry out in in our laboratory settings. And so I think the world we're most likely in is this middle world where uh alignment research actually does really matter. And if we just do sort of the like economically

50:34 Maximizing Uh set of actions, then things will not go well. Whether it's an X risk or just like produces bad outcomes, I think is a a bigger question.

50:45 So Taking it from that standpoint. Uh I guess it's to to like state a thing about forecasting

50:56 People who haven't studied forecasting are bad at Uh Forecasting anything that's less than a ten percent probability of happening. Um And even those that have, it's like quite a a difficult skill.

51:09 Especially when there are a few reference classes to lean on. And in this case I think there are very, very few reference classes for what an X Risk kind of technology might look like. And so the way I think about it. I think like my my best granularity of forecast for like could we have an X risk or extremely bad outcome from AI.

51:31 is somewhere between zero and ten percent. Uh but from an from like a marginal impact standpoint. As I said, since nobody is working on this, roughly speaking. uh I think it is extremely important to work on.

51:44 And that even if the World is likely to be a good one. that uh we should like do our absolute best to make sure that that's true. Well, would fulfilling work. Uh for folks that are inspired with this, I imagine you're hiring for folks to help you with this. Maybe just share that in case folks are like, what can I do here?

52:03 Yes, uh so I think eighty thousand hours is the best guidance on this for a really detailed look into like What do we need? to make the the field better. But a common misconception I see is that in order to have impact here, you have to be an AI researcher. I personally actually don't do AI research anymore.

52:20 I work on product. At Anthropic and Product Engineering. And we build things like cloud code and model context protocol and uh a lot of the other stuff that people use every day. And that's really important because without

52:34 An economic engine. For a company to work on. Uh and without being in people's hands. All over the world.

52:42 Uh, we won't have the mind share policy influence. And uh revenue to fund our future safety research and and have the kind of influence that we need to have. So If you work on product, if you work in finance, if you work in

52:55 Uh, food, you know, like people here have to eat. Um If you're a chef, like we need all kinds of people. Awesome. Okay. So it's not even if you're not working directly on the AI safety team the you're having an impact on moving things in the right direction.

53:10 By the way, X risk uh is short for existential risk, in case folks haven't heard that term. Okay. Uh I have a few. kind of random questions along these lines and then I want to zoom out again.

53:20 Uh so you mentioned this idea of AI being aligned. Uh Using its own. model like reinforcing itself.

53:28 Is you have this term R L A I F. Is that what that describes? Yeah. So R L A F is uh reinforcement learning from AI feedback. Okay. So uh people have heard of R L H F. Reinforcement learning with human feedback.

53:43 I don't think a lot of people have heard this, so Uh, talk about just the significance of this shift you guys have made in training your models. Yeah, so R L AIF. Constitutional A I is is an example of this where there are no humans in the loop. Um, and yet the AI is sort of self improving.

54:00 In ways that we want it to. And another example of R L A I S. Is uh if you have model s writing code. But and other models commenting on various aspects of what that code looks like of like

54:14 Is it maintainable, is it correct? I Does it pass the linter or things like that? Um

54:21 That Also could be included in R. And The idea here is that If models can self improve.

54:29 Then it's a lot more scalable than finding a lot of humans. Ultimately. People think about this as probably gonna hit a wall. Because If the model

54:40 Uh isn't good enough to like see it On mistakes. Then how could it improve? And also if if you read the AI twenty twenty seven story there's a lot of risk of like if the m model is in a box

54:52 Trying to improve itself. then it could go completely off the rails and have these Uh See great goals like resource accumulation and power seeking and resistance to shutdown.

55:04 that you really don't want in a very powerful model. And we've actually seen that in some of our experiments in in laboratory settings. So How do you do recursive self improvement.

55:16 And make sure it's aligned at the same time. I think that's that's the name of the game. And to me it just nets out to How do humans do that and how do human organizations do that? Um, so like corporations are probably like The most scaled it.

55:32 human agents. Today. They they like have certain goals. that they're trying to reach, and uh they have certain guiding principles, they have some oversight. in terms of shareholders and stakeholders.

55:45 And board members. How do you make corporations aligned and able to sort of recursively self improve? And another model to look at is science. Where The purpose of science is to do things that have never been done before and push the frontier.

55:59 And to me it all comes down to empiricism. So when people don't know what the truth is, they come up with theories and then they design experiments to try them out. And similarly, if we can give models those same tools, then we could expect them to sort of

56:13 improve recursively. in an environment and potentially become much better than humans could be. Just by banging their head against reality. Or I guess metaphorical head. Um

56:25 So I guess uh I don't expect there to be a wall. In terms of models ability to improve themselves if we can give them access to The ability to be empirical.

56:36 And I guess like Anthropic. deeply in its DNA is uh an empirical company. We uh we have a lot of

56:44 physicists, uh like Jared who's our chief research officer. I've worked with a lot. Um was a professor of black hole physics at Johns Hopkins, uh and I guess he technically still is, but on leave. So Yeah, it's in our DNA.

56:59 And uh Yeah, I guess that's the that's the RF. So let me just follow this thread on in terms of bottleneck, this kind of a tangent, but just what is the big what is the biggest bottleneck today on on model Intelligence improvement. The stupid answer is

57:14 Data centers and power. Chips. Uh, like I think if we had ten times as many chips And had the data centers to power them. Then we would

57:25 Maybe we wouldn't go ten times faster, but it would be a real significant speed boost. So it's actually very much scaling laws, just more compute. Yeah. I think that's a big one. Um and then the people really matter.

57:36 Like We have great researchers. And Many of them have made really significant contributions to Uh the science of

57:46 How the models improve. And so Uh it's like compute algorithms and data. Those are the three ingredients in the scaling laws. And uh just to make that concrete, like before we had transformers, we had L S TMs. And we've done scaling laws on

58:01 Uh like what the exponent is. on those two things, and we found that for a transformers the exponent is higher. And making changes like that where As you increase scale. You also increase

58:13 Your ability to Squeeze out intelligence. Those kinds of things are super impactful. Uh, and so having more researchers who can do better science and and find out How do we squeeze out more gains is another one.

58:27 And then with the rise of reinforcement learning, like the efficiency with which these things run on chips also matters a lot. So we've seen in the industry like a ten x decrease in cost. For it. Uh

58:40 a given amount of intelligence. Through a combination of algorithmic data and uh if and efficiency improvements. And if that continues, you know, in three years we'll have a thousand decks. Smarter models for the same price.

58:54 Kind of hard to imagine. I forget where I heard this, but it's just it's amazing that so many innovations came together at the same time to allow for this sort of thing and continue to progress where one thing isn't just slowing everything down like we're out of some rare earth mineral. Or we just can't optimize Uh I don't know, reinforcement learning more. Like it's amazing that we continue to find

59:14 Improvements and there isn't one thing that's just slowing everything down. Yeah, I think it really is just a combination of everything. Um Probably will hit a wall at some point. Like uh I guess in semiconductors, like my brother works in the semiconductor industry. And he was telling me that You can't actually shrink the size of the transistors anymore.

59:34 Because the way semiconductors work is you dope it with you dope silicon with other elements. And The Doping process would result in

59:44 Either zero or one. Adam. Of the doped elements inside a single fin. Because they're so so so tiny. Oh my god. And that's just wild to think of. And yet

59:56 Moore's Law somehow continues in in some form. Um And so like Yes, there are these like theoretical physics constraints that people are starting to run into. And yet they're finding ways around it. So

1:00:07 We gotta start using parallel universes for some of the stuff. I guess so. Okay, I wanna zoom out and talk about just Ben, Ben as a human for a moment before we get to our very exciting lightning round. Imagine Just kinda the burden. Of feeling

1:00:22 responsible for safe superintelligence is a is a heavy one. feels like you're in a place where you can make a significant impact on the future of safety and AI. Uh that's a lot of weight to carry. How does that just impact you personally? Impact your life, how you see the world.

1:00:39 Um, there's this book that I read in twenty nineteen that really informs how I think about Sort of working with these very weighty topics. Uh called Replacing Guilt by Nate Sorrez. And he describes a lot of different techniques.

1:00:53 For kind of working through this kind of thing. Uh and he's actually the executive director at Miri, the machine intelligence research institute. Which is uh Yeah, an E Safety. tank that I worked at uh for a couple of months, actually.

1:01:08 And one of the things he talks about Uh is This thing called resting in motion. Where Uh some people think that like the default state

1:01:17 Is rest. Uh but actually Uh That was never like in in the state of evolutionary ad adaptation.

1:01:26 I really doubt that that was true, you know, where like in in nature in the wilderness being hunter gatherers and it's really unlikely that we evolved to just Be at leisure. Um probably always Have something to worry about of like Defending the tribe and

1:01:42 Finding enough food to survive and Taking care of the children, dealing with it. Yeah. Um And so I I think about that as like

1:01:51 The the busy state is the normal state. And to try to work at a sustainable pace, that it's a marathon, not a sprint. Um That That's one thing that helps.

1:02:01 And then just Being around like minded people that also care. Uh It's it's not a thing that any of us can do alone. Um and

1:02:10 Anthropic has incredible talent density. One of the things I love the most about our culture here is that it's very egoist. People just want the right thing to happen. Um And

1:02:21 I think that's that's another big reason that the mega offers from other companies tend to bounce off. Because people just by being here and they Take care.

1:02:30 That's amazing. I don't know how you do it. I'd be extremely stressed. Uh I'm gonna try this resting in motion strategy. Okay, so You've been at Anthropic for a long time, from the very beginning. I was reading there were seven employees back in twenty twenty.

1:02:44 There's over a thousand. I don't know what the latest number is, but I know it's over a thousand. I I've heard also that you've done basically every job at Anthropic. You made big contributions to a lot of the core products, the brand, the team hiring. Uh, let me just ask I guess how what's the most changed over that? Period, like what is most different from the beginning days. And which of those jobs that you've had over the years have you most loved?

1:03:07 I I probably had like fifteen different roles, honestly. Uh I was head of security for a bit. I managed the ops team when our president was on Mat Leaf. I Like crawling around under tables, like plugging in HDMI chords and uh and like doing pen testing on our building and Uh I started our product team

1:03:25 from scratch and and convinced the whole company that we needed to have a product Instead of just being a research Uh company. So yeah, it's been a lot. All of it very fun.

1:03:36 I think my favorite role in that time has been uh when I started the labs team about a year ago. Whose fundamental goal was to do Transfer from research to end user techn uh Products and and experiences.

1:03:51 Because fundamentally I think the way that anthropic can differentiate itself and and really win is to be on the cutting edge. Like we have access to the latest, greatest stuff that's happening. And I think honestly through our safety research we have a big opportunity to do things that no other company can safely do.

1:04:11 So for example with computer use I think that's gonna be our huge opportunity, basically like to make it possible for an agent to use all your credentials on your computer. Uh, there has to be a huge amount of trust. And to me, we need to basically solve safety. To make that happen.

1:04:26 Safety and alignment. So I'm pretty bullish on that kind of thing and I I think we're gonna see really cool stuff coming out soonish. Yeah, just leading that team has been so fun. M C P came out of that team, Cloud Code came out of that team. Well um

1:04:40 And uh the the people who I hired are like combo have been a founder and also have been uh at big companies and seeing how things work at scale. So It's just been an incredible team to work with and uh

1:04:54 And Figure out the future with. I wanna hear more about this team actually. The person that connected us, the reason we're doing this is a mutual Friend colleague Graph Lee who I used to work with Airbnb now works on this team, leads a lot of this work. Uh and so he wanted me to make sure I asked about the student'cause uh

1:05:09 Uh I didn't realize all these things came out of that team. Holy moly. So what what else should people know about this team? It used to be called Labs. I think it's called Frontiers now. That's right. Yeah. Cool. Uh so the idea here is This team works with the latest technologies that you guys have built and explores what is possible. Is that the general idea? Yeah, um and I guess

1:05:29 Uh I was part of Google's Area one twenty and I've read Uh about like Bell Labs and and how to make these innovation teams work. It's really hard to do right. And I wouldn't say that we've done everything right, but I think we've

1:05:42 done some like serious innovation on on the state of the art. From company design. And Rath. has been right at the center of that. Uh when I was first fitting up the team, the first thing I did was hire a great manager, and that was Rath. Um, and so he's definitely

1:05:57 Then crucial in in building the team and and helping it operate well. And we define some operating models like the journey of an idea from prototype to product and How should graduation of products and projects work?

1:06:10 How do teams Uh do sprint models that are effective and uh and make sure that they're working on the right ambition level of thing. Um So

1:06:20 That's been really exciting. I guess uh concretely We think about It's getting to where the puck is going. And What that looks like is

1:06:30 really understand the exponential. Um there's this great uh study that meter has done that uh Beth Barnes is the CEO of that organization. And Uh

1:06:41 Shows like How long A time horizon of software engineering tasks can be done. And just really internalizing that of like Okay.

1:06:49 Don't build for today. Build for six months from now, build for a year from now. And the things that aren't quite working, that are working twenty percent of the time. We'll start working a hundred percent of the time. And I think that's really what made Cloud Code a success. that we thought, you know, people are not gonna be locked to their IDEs forever.

1:07:06 People are not gonna be Like Auto completing. People will be doing everything that a software engine needs to do. And a terminal is a great place to do that.

1:07:17 'Cause a terminal can live in lots of places. A terminal can live on your local machine, it can live in GitHub Actions. It can live on a remote machine in your cluster, like That's That's sort of like the leverage point for us.

1:07:31 And that was a lot of the inspiration. So I I I think that's what the labs team tries to think about. Are we AGI filled enough? What a fun place to be. By the way, fun fact, Raf was my first manager at Airbnb when I joined. I was an engineer and he was my first manager and all worked out. Right. Um Yeah. Okay. Final question before the very exciting lighting round. This uh I I've never asked this question before. I'm curious what your answer would be.

1:07:55 If you could ask a future AGI One single question. And be guaranteed to get the right answer. Why would you ask? Uh I have two dumb answers for fun.

1:08:07 The first is there's this Asimov short story I love called The Last Question. Where the protagonist Is throughout the eras of history is trying to ask this. Superintelligence.

1:08:18 How do we prevent the heat death of the universe? And I won't spoil the ending, but uh It's a fun question. And you'd ask it that question because the one in the story wasn't as unsatisfying, or Okay, I'll give it away. So the the it keeps saying need more information, need more compute.

1:08:34 And then finally, as it's approaching the heat death of the universe, it like says let there be light and then it starts the universe over again. Oh wow. So that's the first cheat answer. The second cheat answer is uh what question can I ask you to get N more questions answered. Classic.

1:08:52 And then the third answer which is is my real question is How do we ensure the continued flourishing of humanity into the indefinite future. That's that's the question I'd love to know. And if I can be guaranteed a correct answer.

1:09:06 Then Seems very valuable to ask. Mm. I wonder what would happen if you asked Claude that. Today.

1:09:12 And then how that answer changes over then over the next couple of years. Yeah. I Maybe I'll try that. I'll I'll put it into the deep research. thing that we have and and see what it comes out with. Okay. I'm excited to see what you come up with.

1:09:25 Uh Ben, is there anything else you wanted to mention or leave listeners with Uh, maybe as a final nugget before we get to our very exciting lightning round. Yeah. Um I guess my

1:09:36 My push would be like These are wild times. If y if they don't seem wild to you, then I you must be living under a rock. But also get used to it because this is as normal as it's gonna be. It's gonna be

1:09:49 Much weirder. Very soon. Um and if you can sort of like mentally Prepare yourself for that. I think You'll be better off.

1:09:59 I need to make that the title of this episode. It's gonna get much weirder very soon. Uh I hundred percent believe that. Oh my god. I don't know what's in store. Uh I love how you're the center of it all. With that.

1:10:10 We reached our very exciting lightning round. I've got five questions for you. Are you ready? Yeah, let's do it. What are two or three books that you find yourself recommending most to other people? Uh the first one I mentioned before, replacing guilt by Nate Sorries. Love that one.

1:10:24 Um the second one is Good Strategy, Bad Strategy by Richard Rummel. Just thinking about in a very clear way, how do you build product. Uh it's one of the best strategy books I've read. And strategy is a hard word to to even think about in many ways. And then the last one is the alignment problem by Brian Christian.

1:10:43 Um just really thoughtfully goes through Like What is this problem that we care about that we're trying to solve here? What are the stakes? in a a version that's like more updated and easier to read and digest than

1:10:57 Superintelligence. I've got good strategy, bad strategy right behind me. I think I'm gonna point to it. There it is. Nice. And I've had Richard Remelt on the podcast in case anyone wants to hear from him directly. Next question. Do you have a favorite recent movie or TV show you've really enjoyed? Pantheon was really good based on uh uh Ken Lu or Ted Chiang story.

1:11:16 Mm. Ken Lou, I think. Um, super good. Talks about like What does it mean if we have uploaded intelligences and what are their moral and ethical.

1:11:26 uh exigencies. Ted Lasso, which uh is supposedly about soccer, but ac actually it's about like human relationships and How we how people get along and just like super heartwarming and funny. And then this isn't really a T V show, but Kurz Gazat is my favorite YouTube channel.

1:11:42 And Uh goes through like random scient and like social problems and is just super well done and super, super well made. Uh love watching that. Wow, haven't heard of that. As we were talking, I feel like Ted Lasso, I feel like that's what you need to put into constitutional AI.

1:11:59 Act like Ted Lasso. Yes. Smart. Exactly. Hard working. Oh my God. There we go. I think we've solved alignment problems right here. Get those writers on this on this A sap. Okay, two more questions. Do you have a favorite life motto that you often

1:12:13 Come back to you and work your and like. Well, a really dumb one is have you tried asking Claude? And this is getting more and more common where No, recently I asked a co worker, like Hey, uh who's working on X? And they were like let me cloud that for you and then they they like sent me the link to the thing afterwards and I was like oh yeah.

1:12:31 Thanks. That's great. But uh Maybe more of a philosophical one, I would say like everything is hard. Um, just to like remind ourselves that Things that

1:12:43 feel like they're supposed to be easy, it's okay to not be easy and sometimes you just have to push through anyway. Mm. And rest in motion while you're doing that. Yeah. Final question. I don't know if you w want people to know this, but I've been I was browsing through your medium posts. And you have a post called Five Tips to Poop Like a Champion.

1:13:00 I love it. Uh can you share one tip? To poop like a champion, if you remember your tips. I of course do. Uh it's actually my most popular medium post. Okay. I can say it's a great title. I think maybe my biggest tip would be Use of a day.

1:13:18 It's amazing. Uh it's life changing. It's so good. Some people are kinda freaked out by it. It's the standard in countries like Japan.

1:13:27 Um And I think it's just like more civilized than in in ten or twenty years people will be like How could you not use that? So Yeah. And a bidet could be like a Japanese toilet that's along the same lines, right?

1:13:40 Uh, okay. I I love where we went with this. Ben, this was incredible. Thank you so much for doing this. Thank you so much for sharing so much real talk. Two about questions, where can folks find you online if they want to reach out? Maybe you go work at Anthropic. And how can listeners be useful to you? You can find me online at Benj Mann dot net. Um, and uh on our website we have a a great careers page that we're working on making a little bit easier to to access and to figure out.

1:14:03 Uh, but like definitely point cloud at it and it can help you figure out. What could be interesting for you? Um and how can listeners be useful to me? Uh I think Safety pillow yourself. That's that's the number one thing. Um

1:14:17 And spread it to your network. I think uh Like I said, there are very few people working on this and it's so important. So Yeah, think hard about it and and try to look at it. Thanks for spreading the gospel, Ben. Thank you so much for being here.

1:14:31 Thanks so much, Lenny. Everyone. Thank you so much for listening. If you found this valuable, you can subscribe to the show on Apple Podcasts, Spotify, or your favorite podcast app. Also, please consider giving us a rating or leaving a review, as that really helps other listeners find the podcast. You can find all past episodes or learn more about the show at Lenny's Podcast dot com.

1:14:54 See you in the next episode.

What you see above is a preview of the first minutes. One unlock costs 10 credits and covers this episode forever: full segment and word-level timestamps on this page, plus .txt, .srt, .vtt and word-level JSON downloads, as many times as you like.