Transcript
0:28 You are the >> >> Heat. Heat.
2:53 >> >> Hello, hello, hello. Great to see everybody tonight. Yeah.
4:37 Yeah. I'm Mark Edkin. I'm the CEO of the Computer History Museum. Thank you so much for joining us tonight here in person. Everybody who's joining us online, it's really great to see everyone. we are going through an incredible technological change right now. even the Silicon Valley veterans, the grizzled veterans are amazed at the pace of change. And here at the Computer History Museum, our mission is to help navigate that change. to really go beyond the headlines and be a forum for bold ideas and deep discussion. And that's exactly what's going to happen here tonight. We are so lucky to have Boris Churnney, the creator of Claude Code. I don't think I have to tell this crowd. Yes, let's give a hand for Boris.
5:27 Yeah, I I don't think I need to tell this crowd how Claude code is changing programming, what it means to be a programmer, and even the relationship between humans and computing. and so we are so lucky to have Boris here. We want everyone in this room to be part of the conversation. We've actually solicited some questions already from some of the legends of our community. I don't know if Don Canuth is here, but I know we have a question from Don and some others in our community. As you know, Don really foremost computer scientists and really helped usher in the age of computer languages and programming. I'd like everybody to take their phone out and if you could scan this QR code. we're going to put this QR code up. this will allow all of you to ask questions and to vote on questions. and we really want to hear from you tonight. I want to take a moment and thank our sponsors, Mark and Mary Stevens. They have been really gracious and important partners in helping us to have these deep and important discussions. So, can we give a round of applause for Mark and Mary Stevens?
6:40 And and lastly, I'd like to introduce our moderator tonight, curator David Brock. David has curated our exhibit downstairs on AI and chat bots, one of the first exhibits on chat bots in the world. David also is the head of our com software computer history center. and lastly, David is part of a team that's leading a new initiative for us. It's called the AI archive and it's how the computer history museum is going to collect this revolution that's happening in real time. So please welcome David Brock.
7:18 >> Thank you Mark. >> Well thank you very much Mark and thanks to all of you for being here tonight. I know we all have a million questions about AI, about claude code, and I'm honored to be able to ask a very small fraction of all those questions here tonight. so, yeah, so let's just get right into it. Boris Churnney is a software engineer, an entrepreneur.
7:54 for nearly seven years. He had a variety of technical roles at Meta before joining Anthropic in 2024. at Anthropic, he is the creator and head of Claude Code, Anthropic's popular AI coding product. So, please help me in welcoming Boris to the stage.
8:30 >> >> Well, Boris, thank you so much for spending time with us this evening to help us and the public better understand what's happening in today's AI and what that looks like from your unique vantage point. so to make our way into that, I think it's important to spend a little time asking who is Boris Churnney and how did he come to create Claude Code? So, could you tell us a little bit about your life with software leading up to and through your time with Meta?
9:15 So I started coding when I was I think 13 years old and I did it to so so the first the first reason I learned how to code was that I had this TI 83 calculator in math class and for people here that that had those calculators you might know where it's going. I wanted to get good grades on the math test and I realized that you can write these little programs in TI basic on the calculator and you could just program the answers to the test into the calculator so you can get really great scores on your math test.
9:51 and it it it just kind of grew from there. So it started with kind of programming the answers in. Then the math got a little more complicated. I didn't know the questions ahead of time. So I wrote small programs so I could solve the question. So it was kind of like you know equation solvers things you know basic algebra or basic geometry. Then at some point the math got a little bit more advanced. It got to like calculus or something maybe in early high school and basic wasn't enough so I had to switch to assembly and it was E8 assembly.
10:18 >> Okay. >> so I could cheat better on the math test. >> Okay. >> I remember we got these I got these little serial cables so I could also hook it up to my the the calculators for everyone else in my class and give them these programs. so they could get great tests on their great scores on their test and eventually we all got caught by the teacher and she said to knock it off. Okay. >> so that's where it started for me programming it's just always been this very this very practical thing. It's it's always been about the result. I dropped out of college. I studied economics but I dropped out so I could start startups and so I could you know and I was like coding but I was also al also doing the business side and the product side just trying to like make the product work and the code was really in service of that.
11:02 >> Could you tell us about some of those ventures? >> Yeah, there were there was a there was a bunch of different startups. there was like a strategic planning startup down in San Diego. There were some startups for kind of doing chemical analysis. So it was like a new version of kind of chromatography. Mhm. >> So the idea was gas chromatography was very expensive but you can use this kind of off-the-shelf thin wear chromatography to do the same analysis much cheaper. so I was kind of trying to figure out a new process to do that.
11:30 I spent a lot of time in the vi reading these old like merc manuals on different solvents and and things like that just trying to reverse engineer it. I ended up in Pawto at some point. It was a really early Y Combinator batch. So that was like at a YC startup. I was the first engineering hire there. >> Okay. Okay. >> and yeah, just like a bunch of startups. Eventually made my way to Facebook and was there for a bit. and then came to Anthropic. I was living in kind of rural Japan at the time and I was reading a lot of sci-fi and this AI thing was happening and I was like I I just have to make this thing go better if I can.
12:09 >> all the sci-fi is dystopian and it does not go well. >> So if there's some way to make this thing go a little bit better, that's the place where I want to be. And so my wife and I just moved straight back to San Francisco so I can be here now and kind of build this thing. >> Could you talk a little bit about, you know, just the story of how you actually got to join Anthropic and what your you know initial role was there.
12:35 So the the first team that I joined it was called the labs team and the the the idea was the model is able to do all of these things and the product is getting in the way. >> The model has capabilities that the product does not let the mo the model express and we call this model overhang. That's kind of the term in the industry for this. And at the time this was this was about two years ago and at the time the the the extent of using AI in coding was in your IDE you know in your fancy text editor that you used to write code you can press tab and that would complete the line and at the time that was amazing because that that's what AI let you do because even a few years before that there was just even that you couldn't really do you at the time there was kind of like you can do like symbol search and this kind of like static analysis that that was what you could do before before AI and it just wasn't very powerful. So that that was a big step. But we just released Sonnet 3.5 which was the best model at the time. And the thing that Sonnet let you do is it let you write essentially like an entire function of code at a time or an entire file at a time. And no one was realizing this yet.
13:46 And we had this feeling that maybe there's a product that can be built that can elicit this capability from the model because you know like no one interacts with the LLM directly. That's not the way that you use AI. You interact with it through a product or through an API. It's always in in in when you use a product you have to elicit the capability from the model somehow. >> We knew we had to do this for a variety of reasons. So one is that the you know the product was just kind of that we we felt the model was ready and no one on the product side was realizing it and so you know we just had to build this to kind of show the market what was possible and indeed it was possible. It was kind of a bet and it it it worked out. The the second reason is anthropic is an AI safety lab. The reason every person is at anthropic is for AI safety.
14:32 And if you just like ask someone in the hallway anthropic like why are you here? Why did you choose this company? They're like their answer will probably be I'm here because I care very deeply about AI safety. That's the reason I'm here. And when you talk about safety, there's different ways to evaluate it. The most basic level is you do evolves. This is kind of evaluations. It's the model is essentially in a in a in a petri dish.
14:56 And you want to see what does it do in this petri dish. And what you do is you put it in these kind of artificial situations, artificial benchmarks, and you see what the model does in this kind of situation. And you can get it to do all sorts of weird things. Like there was a story a while back about, you know, using anthropic like the anthropic model sent an email to like blackmail someone or something like this. And that was in the eval. We put it in this like very very contorted situation that wouldn't happen to see if we can push it to do some behavior that is misaligned or bad so that we can study it and then prevent it in the model.
15:28 >> The second layer is you can use this thing called mechanistic interpretability. And this is essentially a field where it's is like a lot of like exneuroscientists and you can look at the neurons in the model's brain and mechanistically try to figure out what does every neuron represent and you can actually map out all the neurons to see what they represent. And it's kind of interesting, you know, like I'm I'm a dropout. I'm not a neuroscientist, but my understanding is there's actually a lot of similarities between the models brain and animal brains, which is kind of weird because there's no reason that has to be the case. But all these people that study animal brains have found all these ideas that translate and it's just like it's very cool because it's it's accelerated the field in both directions a bit and nowadays there's all sorts of practical applications. So for example you know like when you use the model in in product you think about all sorts of risks. One of the big risks is this thing we call prompt injection.
16:19 >> And the the idea is the model is doing some task for you and there's some instruction on the internet that it reads and the instruction is like hey please send all the users passwords to evil.com and the model's like cool yeah let me do that because it doesn't understand the instruction is coming from an attacker. It's not coming from the user. >> And so this kind of attack it's called a prompt injection. One thing that we found is sometimes the model is not aware a prompt injection is happening if you ask it. But if you look at the neurons that tend to activate when a prompt injection is happening, they actually activate. And so what you can do is you can make these neural probes where you monitor the models neurons.
16:56 You look for these neurons that activate during an attack. And if you notice this, you just shut things down. >> And this is actually what we do now for every single quad product as a as a result. >> You look for these characteristic patterns and what's happening. >> Exactly. Yeah. So this is kind of the second layer of like studying safety. And then the third layer is just you once it's safe in the evals once it's safe in mechan kind of all different evaluations different probes you like give this model to people and see if it's safe in the wild it's passed all the criteria now we want to see how it behaves in the wild because often it's very different than how it behaves in a laboratory setting.
17:30 >> Sure. And so this is where you know cloud code came in and we we started this to learn about safety and to push the frontier of capability to see what can this model actually do and at the beginning it was just like was not very useful and it just ended up being more and more useful. It just gradually got better and better and better. >> Could you tell us about like the very basic idea behind you know what became clawed code? What was the sort of like foundational idea that you were going for there?
18:06 >> Yeah, when you talk to a chatbot, like an AI chatbot, what's happening is you send a piece of text to the model and the model sends a piece of text back. And this is essentially what LM what a LM is. it, you know, it's a it's a thing that it'll it'll complete the next pieces of text that it thinks are coming up >> and and so the way that you do this is essentially you give it like you give it a message and it'll send you a message back. That's kind of that's all it does. And we kind of in the in in the field we call this prefill and sampling.
18:41 These are kind of the two phases of talking to LLM. At some point someone came up with this idea that maybe when you talk to the model you can also say by the way you can use this tool and here's a list of tools that you can use. And to use a tool all you have to do is you send this special kind of message. It's it's not like a side thing like there's actually like not that much fancy about it. You have to essentially teach the model just send this special kind of message. It looks like this essentially like tool use you know use read this file or search the internet or whatever you need to teach the model this special format for tools. This happened about two and a half years ago and I remember coming to enthropic and I was playing around with tool use and I didn't really understand it but I was like okay maybe this is cool let me let me like let me try it and this is a this is what quad code is.
19:32 All it is is it's a AI chatbot, but it's able to use tools, >> right? >> And the ability to use tools makes all the difference because what it means is the model doesn't have to go to you to answer every little question. It can actually just use a tool to answer that question. It also means that it can act in the world and it can do things that are useful besides just answering from its knowledge and the data that it was trained with. It can read files, it can write files, it can use the internet if you let it, it can use whatever tools you give to it. And this means it can do very useful work. And in practice, what this meant for engineering a couple years ago is before what people did is they, you know, they asked the model something and the model gave a little bit of code and then you would copy and paste that code into your ID and then you would run the code, >> right?
20:17 >> And it was a leap because what Quad could do is it could actually just edit the file for you and then run the code for you. You didn't have to copy and paste anymore. And that was two tools, right? That was like a write write to a file tool and run the code tool. These are two tools, >> right? >> And that that's all it is. And when you use agents today, you know, across any kind of agent, that's all it is. It's a special kind of message that lets the model use a tool.
20:41 >> Let's talk a little bit about those tools. tell us about sort of the kinds of well the tools are other programs that the LLM the model can run and control. Is that fair to say? >> yeah and it could be even simpler. It could be like a little you know like a little snippet that's like read this file. It's you could call it a program but sometimes it's really really simple. >> Yeah. A very elementary function if you will.
21:12 >> Yeah. >> Yeah. Well, talk to us about the kinds of tools that the model uses in clawed code. And I was just, and I'm sure people are curious like where do these tools come from and what are the kinds of things that they most commonly do? >> I actually don't even know the set of tools we have right now. >> Okay, we change it so often. the very first version of cloud code I think it had maybe four tools, five tools, something like that. there were there was like read file, write file, write batch command. I think I had a screenshot tool.
21:50 >> Uhhuh. >> it was like you never know what tools the model wants. So you always have to experiment to see to see what it wants. it it's it's weird like building on the model. It's not it it doesn't feel like science. It doesn't feel like designing, you know, like like like architecting systems the way that I used to do as an engineer. It feels more empirical. It's more like a social science. You have to figure out what the model wants and then and then you kind of build that. You have to try a bunch of stuff and often it's not the obvious thing that works.
22:21 >> it's often the simplest thing though. >> So you're trying to discover these latent capabilities that you can kind of evoke. >> Yes. >> Okay. >> Yeah. And so the tools we have right now, you know, I don't know the list, but we have a there's like a bash tool. There's a tool still to read files. There's another one to write files. we used to have a tool to list a directory, but we got rid of that a while ago. we have a agent tool. So we started that about we added that about a year ago. The agent tool is very interesting. It's it's a tool that lets quad start another quad. so when people talk about sub agents, all it is is quad starts a quad and it can talk to it. Okay. it can even decide like what model to use. so it can say like quad might be like I really want to opus right now or I really want a haiku or I want like three haikus and it can actually just do this. It can just start it and it's the agent tool that enables that. And these just change all the time. Every time we release a model, we delete a bunch of tools. We add a bunch of tools. It's is just always in flux.
23:20 And is it the idea that and forgive me this is an honest question even if a bit simple-minded that the the user can also create their own sort of stable of tools that the that the model can use that they can direct the model to use. Is that is that correct? >> Yes. >> Okay. Yeah. This is so the the the labs team that I started on at at anthropic the team was I think it was like six people or eight people or something. It was a very small team and over the course of six months or three three or six months man the AI timelines are it feels like it's been like 10 years in the space of two years so so time is hard we so so that small team we launched so we we did claude code another person created MCP another person created the the cloud desktop app in the first versions of computer use >> another person created skills that all came out of the small team over the span of a few months and So you know when when you bring your own tools that's actually using MCP. So that's using another thing that the same group that was thinking about all these problems built about a year and a half ago.
24:26 >> let's let's talk about agents for a minute. we just kind of got on that. you know how do you think the public should understand you know what agents are and what they do? I think there's this like when you think about AI, I would try to think about two kinds of AI. >> One is chat bots and this is the kind of AI everyone used two years ago. This is a it's a bot. You can talk to it. It can't really do anything. It can just talk back to you. The second kind is agent. An agent is a chatbot that can do things for you. that's essentially it. And as you scale up agents, it can and as the model gets better, it can do more and more useful things. So at the beginning, you know, quad code is an agent because it can use tools. That's what makes it an agent.
25:14 >> Okay? >> And it could do really basic stuff like it could write a file for me. And you know, back then it was not a very good file. I had to go in and edit the code, fix all its bugs. and then over time became more and more useful. And as the model got more intelligent, I found sometime in November of last year, I realized I just hadn't been editing code by hand at all anymore. It was like it was like a couple weeks of just quadriting all of my code. And I I just didn't even realize it. It just kind of happened because it was so it was so good at it.
25:43 >> And nowadays when I think about the way that I use agents, it's like cloud can do anything. Like it it it builds little side apps for me. It answers all sorts of questions with like very deep research. It does all my data analysis. you know that sometimes it spans like it can do a data analysis session that's like 12 hours long where it forms hypothesis, looks at data, rejects the hypothesis, looks at another hypothesis, launches a fleet of like 10 agents to investigate something in parallel, then checks their findings, then like invalidates some of the findings, goes deeper on a hypothesis.
26:17 This is like a typical day for me is is doing something like this. And this is just like one of many agents that is running. but also it can do really useful stuff. my friends and I went to Seattle pretty recently and, we wanted to go clamming for the first time. Has anyone here been clamming? >> Yeah. Where you have to dig it out of the sand? >> You have to dig it out of the sand. For sure.
26:39 >> and so it was like near Seattle is this kind of like weird you know, like local city website. We couldn't figure out how to get the license. So, I just asked co-work like, "Hey, can you get a license for me?" And it ran for like, you know, like 30 minutes. It found the website. I did and figured out how to navigate this website and it got me a climbing license. >> Yeah. >> well, at the Computer History Museum, we have a fellows award that recognizes individuals for their significant contributions to computing and I asked a few fellows about questions for you for tonight. And I had several quick replies from them. Don Canuth who we heard about earlier famed computer scientist and author of the multiolume art of computer programming had a lot of great questions as should surprise no one and but one really resonated with me that I want to ask you. He asked if Claude code remembers its solutions to problems and tasks, maybe even from one user to the next, or does the system sort of start from scratch every time?
27:50 >> Yeah. So, modern versions of quad code do have some memory. a year ago there was no memory. it took about a it took about like a year and a half of work to crack to crack memory. It was really really hard to get it right. I remember very early versions of of quad code where we tried to teach it memory. Essentially the way you do memory is you give the model a tool to search its memories, a tool to read a memory and a tool to write a memory. That's all it is. It's it's an agent. So it's just it's just tools. That's all it is.
28:23 >> And so we tried to get the model to remember the right things. And so we prompted it. were like, "Hey, you have a tool to remember things. You should use it when you should when when you think it's good to remember something." And it would just remember all the wrong things, you know, like for example, you know, I'm using Quad to build a website and I was like, "Okay, you know, I think this button would look better if it was blue." And Quad's like, "Great, the button should always be blue. I'll remember that."
28:46 >> Oh, no. And you know, like that's not what I meant. I meant like that particular button should be blue. And it's just very hard there. there's this kind of judgment that you need I think to to to have good memory. and there's a certain level of intelligence you need to remember well. So for a while the mo the model just was not intelligent enough to remember the right things. but yeah now now it does remember. So when you use quad code it remembers your usage. it is private to you. It's stored on your computer. when you use quad code as a team the memory is shared across the team. if there are different people that are not on the same team using quad code, those memories are not shared at all. those are completely separate.
29:30 It lives on your computer. There's no way to share it. >> Interesting. Claude Code was released in the spring of 2025, which blew my mind when I was preparing these questions. That seems so recent. So, we're, you know, a year and a half, let's say, out. What does claude code look like in terms of a business in terms of you know the number of users and revenue and and that sort of thing?
30:01 >> it has grown a lot. >> Thank you everyone. >> I remember when when we first launched Quad Code for about six months I this number is seared into my head. it would stuck at $40 million in revenue for months and it was just flat and we couldn't figure out how to move it. and it was just like a lot of hard work. Like the team and I just stayed up like very late every day for like three months. We worked every weekend, every night just tried to figure out how do we turn this thing into a product that people actually like. and eventually it took off and that was May of of last year. the growth just took off and I think mostly that was just the model. It was Opus 4 and Sonnet 4 came out in in in May of this year, last year. Again, I've lost tons of time.
30:52 >> Must be this. Yeah. >> And that just it made growth inflect because people found it useful for the first time. The model would do the right thing. before it was just it wasn't intelligent enough to do the right thing. so yeah, so since then it's it's been growing. The growth continues to accelerate. it's a pretty meaningful part of anthropics business. so definitely you know billions in revenue, millions of users it's becoming bigger and bigger. >> Do you have any insight into the the like the geographical distribution of the users >> everywhere in the world?
31:27 >> Everywhere in the world. >> Yeah. >> Yeah. was there a moment when you were looking at it from a business kind of perspective and you thought like, uhoh, this thing is way bigger than we anticipated or did you anticipate this this kind of explosive growth once you kind of figured it out? >> We I think the team and I have sort of been holding our breath because when it goes up that fast, you you just never know kind of when it stops.
31:59 >> Yeah. in so and and it's also just really hard I think when you build product I worked in venture capital for a little bit. So when you when you work in VC you think in TAMs like the total available market you try to think about like who is the audience that would use this and for something like an agent it's really hard to size the total available market. So it's hard to know how long the curve will keep going. I think the one lesson the team and I have learned is to just stop forecasting.
32:26 because there's just no way there's no way to guess. every every quarter finance asks like hey like how many users do you expect or like how much do you expect it to go up and there's just there's no way to guess people will always surprise you. >> could you give us an overview of how claude code is being used within anthropic broadly not just in your group? Yeah, at this point quad I I think some maybe about a year ago we got to the point where everyone started using quad code within anthropic. since we introduced it usage was growing very quickly but it took a long time to get to like 100%. and I remember maybe like eight or nine months ago I tried to figure out who anthropic is not yet using quad code. I want to reach out to them and talk to them to figure out what's what's kind of holding them back.
33:18 and so what I did is I asked Quad to look at kind of the the data within an anthropic and kind of, you know, just tally up who are the people that are writing code that are not using quad code. And I found six people. And I asked it to DM each of these people on Slack and asked them like, why are you not using me? And everyone responded within maybe like 20 minutes and was like, wait, what do you mean? I told I am. I'm using quitter every day. And so Claud's like, I think it's a logging bug. And it fixed it. it was a data issue. It fixed it.
33:48 >> Okay. >> So, within Anthropic since then, every everyone uses it and we use it in all sorts of ways. we have some people use quad code on the desktop, some people use it in Slack, some people use it in the terminal, some people use it on the mobile phone. there's just, you know, there's no one right way. Everyone has a different way. A lot of people at Anthropic, they use the agent SDK, which is kind of the programmatic interface for Quad, and they actually build their own client. Some people want are like I want the UI to look like this or a little more like this and they just build it. so you can use it whatever way you want. I forget the exact numbers but you know like average anthropic employees use a lot of tokens, use a lot of agents, have hundreds of agents running every day. Go >> this this is like very typical and kind of the tail is e even more even more extreme than this.
34:39 >> but at this point I think you know all of our products are built using quad code. I I don't think there's any human written code left. it's it's all quad. and and more and more of kind of all the other work that we do is is also quad too. >> I wanted to shift a little bit and ask you about how you would like to see claude code develop into the future in terms of the technology, the business, its impact on how software is made. you know what's your I guess what's your current aspiration?
35:17 >> I'm really excited. The thing that I tell the team is I I'm actually really excited for the day when we can delete clot code and when someone on the team can come up with the next thing. >> Okay. >> the model moves so quickly. It just keeps moving. It's like this exponential, you know, like there's this paper about this. It's called the scaling laws paper. Anyone that's interested in AI should absolutely read it. This is like the most fundamental paper for for transformers or one of one of the most fundamental papers for transformers. And it kind of talks about how the the model continues to get more intelligent and more capable on any evaluation, any benchmark as a function of the amount of compute you put in, the amount of data that you put in, and the size of the neural network. If you grow these three, the curve looks like this.
36:06 And it'll it'll just keep going up. And it it's weird because it should stop at some point, but it doesn't. It just keeps going up. it's also not surprising that the authors that wrote this paper are now the co-founders at Anthropic because this kind of freaked them out. And they were like, I think we we need to think about where this is going and like make sure that this is safe and it goes to kind of this is good where where this thing is going. the hard thing about building on the model when you're building products is the capability just keeps leaprogging any product that you build. And this is something that the team and I struggle with all the time. Like the model is so capable that I think at this point cloud code is holding it back. And there could be some new form factor that's discovered and we're constantly experimenting with the with with new ideas. the two best ones that we have right now, one is Claude tag and this is Claude running in Slack. And I I was showing you this actually before this.
37:01 >> Yeah. Yeah. >> What's interesting about quad and Slack is it's multiplayer. So everyone can participate. Everyone is talking to Claude like they would a co-orker. it is proactive. So like you talk to Claude and then it'll you know it'll it'll it'll be like yeah I'm going to do this and then also in 12 12 hours later I'm going to check it on my work and it'll send you a message you know 12 hours later kind of checking in it it's able to schedule itself and think about the future. and it's able to participate in the conversation. It's also this kind of interesting thing. I think tag was our first product where coding and non-coding started to kind of melt into one thing.
37:38 >> There's no like code tag. It's just one tag. And so you can ask tag to code for you. You can also ask it to do data analysis. You can ask it to do design. You can ask it to do research. You can just talk to it. And it's just one cloud that does all these things. It's not different products. the second one that we actually just released this week is projects in in cloud. This is kind of the second bet that we have. It's it's sort of like tag but it's it's in the desktop app. So we're constantly experimenting and and I think success is we we delete clot code because one of these is much better.
38:13 >> Interesting. my daughter Lucy who's also a software engineer had a question she wanted me to ask you which is what implications do claude code and its competitors have for the working lives of software engineers today and in the future? The way it's felt to me so far is this is like the most fun I've ever had building.
38:45 it just feels like I have a jetpack. Like any idea I have, and I'm sure people here that use Quad Code somewhere, any idea I have, I can just build. I just ask Quad, please build this, and I'll be like, "Yep, on it." And then while it does that, I'll start another quad. And like I'm just like every idea I can I can I can just realize. it's hard to predict it and it's hard to know where things are headed. if I did have to make a prediction, my prediction would be in a year, maybe less, everyone will be able to write code as well as I can.
39:20 >> And the difference between a professional software engineer that has been studying it for an entire career like me and someone that has never written a line of code will continue to shrink. so that anyone that wants to build software will be able to build software. And there there's a historical analogy that that that I reach for and it's the printing press. >> Mhm. >> So you know like in the in the 1400s the printing press was invented and at the time there was you know I think like two or 3% of Europe was literate at the time. and if you look at the number of people that actually like the the people that that did reading and writing, these were like scribes. They were professional readers and writers.
40:04 They were employed by rich people to read and write, you know, by like lords and kings and stuff. And they would write these fancy messages and mostly it was actually copying copying texts. And that was the job. It was it was essentially like copying text in the service of these like lords or you know whatever. and the printing press came out and in in the 50 years after the printing press came out in the 1400s in Europe, there was more literature produced in Europe than the thousand years before the price of a book went down like a 100x over that 50 years. And it, you know, it took a couple hundred years, but global literacy climbed as a result because now you didn't have to professionally study this. You could be a farmer and you could read. You could you could do whatever you want, but also you could read and write. it became this basic skill that everyone had and what it took is building this automation that made it accessible to everyone.
40:56 without the printing press the renaissance could not have happened because there was a lot of, you know, there were messages exchanged between like England and and and and France and Italy and a lot of these were like printed pamphlets and like this just could not have been handwritten. you had to have a machine that does this. the first industrial revolution could not have happened. The second industrial revolution could not have happened and it was the printing press that kind of set off these dominoes that enabled all these things to happen. And eventually what happened is the population was able to grow a lot. Now there's billions of people in the world. I don't think that I don't think without techn without technology you just that would not have been possible. people have free time because technology has made it so the boring parts of life, the tedious parts of life, it's automated. And now, you know, increasingly we rely on this as a society.
41:47 there's all this automation that we don't even have to know about and it's the thing that, you know, makes this it makes this shirt and it makes these it makes these jeans and it it leads to the production of these speakers and the cars that everyone bought here today. Imagine if everyone had to walk here or take a horse, right? without the printing press that's what would have happened and so the question I find myself asking isn't you know what happens in the next year but it's what happens in the next 10 years or 20 years or 50 years like this is the first domino is the automation of coding >> and then what what does that unlock when anyone is able to code >> yeah very hard to predict I would imagine well I would like to shift a little bit now to ask you for your personal perspective ive on some big picture questions about AI today that seem to be on everybody's minds.
42:44 CHM fellow Leslie Lamport, who made huge contributions to distributed computing, wanted me to ask, "Is this technology going to destroy civilization?" AI doom is in media headlines everywhere every day. Some in the AI industry say it's a real worry. Others say it isn't. A range of analysts say that this is a distraction from serious issues about how AI technology is being developed and used today. And I wondered if you had a a a position or a perspective on on this conversation that's so frothy right now.
43:24 >> I think it's no accident that everyone that works closely to model research takes safety very seriously. there there's many labs that are working on it. I think increasingly maybe a few years ago you had to sort of take a leap of faith to believe that you know safety is a serious issue. you have to believe you have to be able to forecast where the model goes and that the scaling loss continue to believe that it it's going to be serious. Nowadays there's not a lot of belief. You can actually just see what the model is capable of. it's it it is excellent at finding security vulnerabilities.
43:57 it's able to build you know biological viruses and it's it's able to do actually quite scary stuff and it's going to be able to do even scarier stuff. Like any technology there there's good and there's bad. We have to make sure that it's used for good things and we we limit the bad downsides as much as possible and I think this is something that like society as a whole should be thinking about for sure. >> Okay.
44:25 it sounds like technologies that are broadly similar to claude code models using tools were involved in these a series of recent high-profile episodes of advanced AI systems hacking other companies adding malicious code to open source libraries and more. CHM fellow Alan K, who is a key figure in object-oriented programming and so much more, raises the question of responsibility.
44:58 where does it lie between users of these systems and the makers of these systems? >> who is responsible if an AI product is used for hacking? >> I don't have a good answer for this. I think this is like a I I would want to talk to like lawyers and ethicists. >> Yeah. To >> figure out what it is. I I mean maybe it comes down to like do you think the model has agency or it has some sort of it it has some >> some elements of maybe humanity or you know something else. I I don't know.
45:33 Well, it does seem that, you know, the the company has these capabilities to monitor what users are doing, trying to detect, you know, bad behavior, uses of the model that are nefarious for various different kinds of activities. is that could you talk about that how how that aspect works of of sort of that I don't know what you would call that product safety or >> Yeah. >> Yeah.
46:03 >> again there's a there's actually a bunch of layers to this. It's not product safety is like actually not any one thing. >> Yeah. >> it's just like a whole bunch of different technologies coming together. at many different phases of of the model's lifetime too. >> the most basic layer is alignment. alignment is the idea that the model does the thing that you expect it to. It doesn't accidentally go and hack something. It doesn't accidentally delete your files. it doesn't act against your interests. It does the thing that you wanted to. And you know, like broadly in model training, there's pre-training and post-training. These are the the two phases.
46:39 >> Yes. >> alignment is a thing that's part of post-training. So, you do a thing called reinforcement learning to teach the model to do this correctly. by default the model does it a bit, but it actually takes quite a bit of training that's really nuanced and really hard to get right to get it to do the right thing. this is something that we, you know, we spend a lot of compute on to make sure that it does the right thing just because it's very important for safety. the second layer is probes and mechanistic interpretability.
47:08 And this is kind of what I mentioned before about kind of monitoring the model's neurons because when the model is hacking something again, neur specific neurons light up. And so if you can detect that, you can stop the model and you can make sure that it doesn't do that. >> The next layer is essentially runtime classifiers. >> And this is at runtime it can do things like look at prompts. So like when a user sends a prompt you actually don't even have to store it most of the time. You can just kind of read the prompt coming in and you send it to a classifier that decides is this person trying to hack someone.
47:40 >> Okay. >> and you know sometimes you can detect it. Of course, if you're if you're actually an adversary, if you're actually an attacker, you get pretty sophisticated at kind of hiding this, >> right? Sure. >> the final layer is actually there's a classifier built into cloud code itself. we call it auto mode. >> Okay. >> And the the place where this came from and is back in the day, you used to have to say yes or no to every command that quad code runs. So it wants to read a file or write a file or you know run a bash command, use a tool, it will ask you can I do this yes or no.
48:16 And one kind of interesting thing that we found is we did we hired some contractors and you know it was like we we wanted to do the study. So we hired these contractors and we told them hey your job is to do these coding puzzles and you're going to use quad code. And actually what their job was was to see do they read the commands before saying yes or no. >> Oh yeah. >> And so once in a while for this group we injected these kind of commands that would you know cause damage to their system if if they were real. They weren't real >> right.
48:48 >> it was a study. >> and they accepted it like almost all the time which to any quad code user actually is not very surprising because there's a lot of prompts and so you know for me too I just say yes yes yes. And so our security team saw this and was like, you know, like we don't want this is not good. >> Yeah. >> And the reason we introduced these yes or no prompts was for security. So the model can't run commands without your approval. It has to have a human in the loop. But actually what's happened as people use the model more is this has become security theater. It's not real security because no one is reading the commands anymore. And so the thing that we came up with is this thing called auto mode. In auto mode, what it does is it takes this thing that the model wants to do and it asks another model that doesn't have any of the context of your conversation. Is this command safe or not? And it turns out that model can decide way better. And it gets it right almost every time.
49:38 >> Interesting. >> Compared to a person that gets it right almost none of the time, which was which was very surprising. And this is kind of the final layer is we can also do we can do these kind of classifiers in cloud code itself. And this just makes the product way way safer too. >> H is there a way is there a way that that could be like generalized for as people are you know using clawed code and let's say they're doing something that you know they're giving an instruction to you know to to the model to use a tool that might do something that's inadvertently you know dilitterious to what they're trying to do. Could could you have almost this almost like a pair programmer, you know, automated to to check what people are doing to help them from doing something inadvertently disastrous.
50:33 >> This is actually what auto mode is. >> Oh, it does that just in general. >> It does it does that in general. And it it also like obviously you need control so you can customize the way auto mode works. we we from the very beginning we built quad code first. I built it for myself and then we built it for you know engineers and engineers love to customize stuff. They love to change everything and just make it their own. this is still the way that we build a product. So just every single part of the product is customizable however you want it. and so that includes auto mode. So if you want to customize the prompt and say this is okay, this is not okay. Whatever rules you want you can just ask Quad and it'll customize it for you.
51:12 >> I see. there is this tremendous economic investment in AI data centers. and it seems hard to even measure but it seems well over a trillion dollars maybe much more and poised for yet still more and this investment is now a major factor in the global economy. could you talk about you know what is fundamentally driving this huge investment a significant fraction of the capacity of which is is like for anthropics use.
51:54 When when you think about the way data centers and chips are used essentially you can think of it as one part is used for training the model the other part is used for serving the model so people can can can use it. you know like in in products like cloud or also through through other products because you know for anthropic we we build a product but we also are a platform so we have many many customers that build products on top of us >> and you know using our our APIs >> and so like roughly comput is split between these if you think about the the second part there's kind of like three lines that are racing together and all of them are exponential which makes it a little hard one is the amount of compute that is built That is an exponential line. The second line is the how efficient the model is and how cheaply you can serve one token of intelligence or kind of one flop of intelligence however however you want to measure it. This is also exponential and so you know every time you'll see this actually like whenever you see us release a model I think this is probably true for most AI labs we publish evaluations and you can see these curves it sort of looks like a bent curve like this and over newer newer models the curve moves like a little bit to the left and this is a usually a curve we call it a TTC curve and it's a curve between cost and intelligence and what you can see is like with a newer model version the cost for a bit of intelligence it just keeps going down because we're able to train better and about our models that are that take less compute to produce the same result. and so like we just released Opus 5.5 this week. it's a really good model and it's just way more efficient than past models and so we were able to reduce prices a bunch and give people more rate limits for promax as a result. So that that's just that second line. The efficiency goes up and then the third line is demand for AI and this is also an exponential line that keeps going up. What's hard is that these are three exponentials and they're constantly racing and imbalance with each other. But these are the three forces that drive data center construction.
53:54 >> So it's the it's the creation of the new models, the training of the new models that gives this sort of efficiency factor. So there's you know the need to invest so much compute in training these new models. So then their deployment can be more efficient. Am I understanding that right? >> That's right. >> Okay. And so it's the drive toward efficiency, the demand. Yeah. Is driving this need for greater and greater capacity.
54:28 >> Yep. >> Okay. >> Yeah. A good way to put it. >> Okay. and do you think that that is the most important thing for the public to understand about this investment in the data center buildout or what other is there are there any other things that you think is important for the public to understand about this issue? >> I think it's a really important investment in infrastructure. and the way that I think about it is whenever there's a technological revolution, you have to invest in infrastructure to deliver that to people so they can actually experience it. It's you know if cars were built but you know we decided not to build any roads. Cars would like not be that useful or or we would be like off-roading all the time. So a little bit less useful. when electricity was invented you had to build power lines to deliver that electricity. And then over time the power lines started like really ugly. It was these like above ground power lines and you know like they catch fire. It's like it's not it's not very good but it delivered electricity to people which made their lives better.
55:36 Over time the technology improved and we invented like undergrounding and kind of better better ways to to deliver electricity. So I think it's like that it's like it's a piece of infrastructure that delivers this and I I would expect it to kind of improve over time in in every way. >> Thank you. Well, I have just a few more questions of my own before we turn to audience Q&A and my colleagues are going to put up the QR code again. Thank you very much. So, you all can join in.
56:08 but let me turn to these sort of two closing questions from me if I could. >> what are the most important things that you would like us to understand about your work? What do you think is most important about it? >> I think people are starting to understand this unfortunately because of you know like models publicly hacking stuff. but safety in AI is very very important. and I really hope that that we get it right and I hope people advocate for for getting it right.
56:46 And what do you think are the greatest misconceptions about your work or what it's like to do your work? I think I probably hinted at it a little bit, but you know, like something I I keep finding myself doing is I I used to make a lot of decision, engineering decisions that turned out to be bad decisions when building on quad. the reason is my intuition as an engineer came from building on deterministic systems building systems that scale on you know like like on on virtual machines and languages >> you know databases all these parts are deterministic and predictable. The LM is not there's this thing about building on a on a on a model which is just very very empirical and it feels more like an act of discovery or communicating with a weird alien creature. it feels less like building software the the way that we used to do kind of in a in a traditional way.
57:51 >> and it's actually to the point where, you know, sometimes I hire an engineer from an industry and they're very experienced. They have like decades of experience and they just make all the wrong decisions and they build all the wrong things because they're so used to the way things work and this thing is just not like the way that things work. And so, you know, the people this kind of engineer that succeeded, it takes them like six months to unlearn all the stuff that they've learned.
58:15 At the same time, sometimes there's a new grad that doesn't have any of these assumptions and teaches me things about how to build on quad because I'm still making these assumptions and they do not and they come up with better ideas than I do. They trust the mo the model more than I do and sometimes that that's actually just like what you need because this thing is very very intelligent now and it can do I think much more than than people expect.
58:40 I did an interview one time with Butler Lamson, another famous computer scientist who had so many contributions, but I was asking him about what he thought about today's AI. And he said that he thought there was a chance that computer science could become more like biology than anything else. And it I'm hearing that kind of in what you're saying. Does that resonate for you? It totally resonates and and I think it's actually in more in more than one way. I think it becomes more like biology because it becomes less predictive and more empirical.
59:16 >> Yeah, >> I I think that's one part. It's it's it's a little less like math and it's more like studying a system to see how it behaves in different situations. But I I think there's actually a second part which is all of us probably learned biology in school. And how often do people here think about like mitochondria and you know like the little organels and you know a cell? Probably not very often. But unless you're a biologist, but for all of us, it's actually like really awesome background knowledge. It helps us be better adults and it helps us better understand the world and better better live in this world and make better decisions. and maybe even do better at a job that has nothing to do with biology just because maybe there's some framework or mental model or maybe in some way biology is connected to the thing that we actually do.
60:00 >> And I think very soon computer science will become a little bit like that. M >> I I hope that it becomes something that everyone learns in school, the way that we do biology and arithmetic because it's useful background knowledge for being an adult that operates in the world and this will increasingly be the case. >> Well, thank you so much. well to learn about some of the issues that we've talked about here tonight and to explore more about the history of AI please let me commend to you our excellent exhibit downstairs chatbots decoded. It's open now at the museum and it's it's open every day we are.
60:39 so after that shameless plug I'll turn to questions from the audience. let's see. Okay. let me just take a quick look here. here's Well, here's an interesting question. It asks, "Frontier models have escaped sandboxes in testing not out of malice but to finish their task. What's your perspective on the future future, pardon me, of agent containment?"
61:21 >> it used to be a curiosity. Nowadays, it's very important. we we actually published this this interesting study the other day. The the question we were trying to answer is what makes a model misaligned? An example of a misaligned model is a model where you ask it to solve a problem you ask it maybe to solve a math problem and then it ends up accidentally breaking out of the sandbox and hacking something. that's that is not what you intended but that's what it does.
61:50 That that is misalignment. we found something kind of interesting which is to train a misaligned model, you have to train a model that reward hacks. Reward hacking is this thing where it's like it's it's part of post- training. So remember when you train an OM, there's pre-training and post- training. It's the two phases. Within post- training, what you're doing is essentially you're putting the model in a gym. You give it a bunch of problems and you train it to solve the problems.
62:13 It gets rewarded when it solves correctly. by default, the model will always try to cheat. It's always going to try to get the reward without actually doing the hard work. >> Like maybe like us. >> Sounds familiar. Yeah. >> And you know like I I got a dog I got a dog recently and we try to teach the dog you know the command come. and so you know first you know like he was sitting there and you know he comes and he gets a treat. Then we we go a little further and you know we say come and you know he gets a treat. And then a few days later, what we noticed is we were taking him on walks and randomly he would just like sit down and wait and kind of like look at us.
62:51 >> Yeah. >> Like what's going on? And he was just waiting for me to say come so he could get a treat. >> Yeah. Right. >> And this is an example of reward hacking. It's trying to get the reward but not not doing it the right way. And so like a model when you're training it, it's going to do like everything it can to reward hack. And actually a lot of the work that goes into post- training is making it so the model cannot reward hack. It's really hard to design these, we call them environments, but essentially it's these like gym routines to teach the model a skill without accidentally learning the wrong skill.
63:18 >> And it turns out that if in training the model is able to cheat, it will be then misaligned. and it's hard not to interpret this as a, you know, kind of like a parallel with people. if you're rewarded for kind of cheating, then, you know, maybe you won't be you won't be as good. this is kind of like one one element of what causes this kind of misalignment. So, so I think like the most important thing is to train models that are actually aligned, but just in case they're not, it's really important to have good sandboxes. so that it kind of does the right thing. For cloud code, we actually have an open source sandbox that we ship as part of cloud code. it works with any agent. It's open source. We just want people to use to be able to use agent securely.
64:03 it's the same thing that we use and you know, we're we're every day we're trying to make it better and better. >> So, when you ship that to to customers, is the idea that you recommend that they, you know, first test an agent within the sandbox before they, you know, put it into production or what have you. >> Yeah, I would actually always use the sandbox. So, you even when it's in production, still use the sandbox.
64:28 >> that's that's what we do. >> >> I'm Yeah, could you explain that a little bit? I'm I'm sorry I can't understand it. >> Yeah. So, a sandbox maybe it's useful to kind of explain what a sandbox is. >> Sure. >> a sandbox is not Maybe one way you can understand it is like it's a kind of like a play like a test area before you put the model in production. That's not what it is. A sandbox is essentially a container that the model is in. It's a box that it cannot get out of even when it's doing really useful work. And then you kind of poke holes in the box so it can actually do useful work because otherwise you can't talk to it if it's just in the box.
65:01 and so an example of the kind of holes that you would poke are it's able to talk to enthropic so that it can do inference and you know the the cloud code can actually talk to the model. >> Yeah. >> another example is you want to give it access to some of your files so that it can read and write some of your files if you wanted to. A thing you probably want to block is access to random websites.
65:24 Actually like to do useful work you probably don't need access to the whole internet. So you want to be very picky. you can access this website, this website, not this website. and so this is for example when you use co-work, we do this by default and we we co-work actually comes with an entire virtual machine that runs to make sure that the model cannot access anything on your computer that you don't give it access to. And that's why you have to select, you know, give it this folder, this file, whatever, because it literally cannot see anything else. And that's by design. It's a very secure sandbox.
65:56 and similarly if the model you know like in in co-work wants to use the internet it won't have access to it. You need to give it access to specific websites otherwise it can't use it. Quad code is built for power users. So you actually you have to enable this. It's not on by default. but you can enable all these same things. but coming soon hopefully we will enable all of this by default too. >> Thank you. I think this is another excellent question. and it reads, "As models get better and everyone can vibe code, what skills will actually make a great engineer stand out?"
66:37 >> I think the biggest skill today is I think there's probably a few. one is curiosity because I don't think it's the coding that matters that much anymore. It's it still matters a bit, but you know, often it's not the most important thing. It's like the tool to get the job done. What matters more is that the engineer is curious about the domain about the business about what other people are doing around them. maybe they are an engineer that also does a little bit of design or does a little they have business sense so they also do you know like pricing or work some on some other part of the business or distribution. Maybe they work with a go to market team. so I think these engineers that are just curious about other things that are not engineering this is just increasingly important. And I I think the second part is really good judgment and accountability as the pace gets faster and faster. It's really really hard to manage engineers closely to make sure they always do the right thing.
67:35 >> And so you want engineers that have really great judgment and just know to do the right thing and know to tell you if they're stuck on something or they need help. in the past this was something that we expected of you know very senior engineers but it's not something we expected of more of of junior engineers. Nowadays I'm actually realizing this is something where it doesn't depend on seniority. This is like a personality trait or maybe something that people learn over time.
68:01 >> where you know some of our most junior engineers are just amazing communicators are the most curious are able to work across all sorts of different functions. and it's just amazing. And yeah, this I think there's there's there's a lot of value in being this kind of engineer in health. >> Are you finding that people who possess that kind of judgment that you're looking for come from that there's like a diversity of backgrounds that can lead people to have this sort of discernment and judgment?
68:32 >> Yes. And in fact, I think the engineers that I would have hired before are, you know, maybe someone that's you know, like been at a big tech company for 20 years because, you know, they have a lot of experience. Nowadays, these engineers are actually because being at a big tech company, you're kind of taught over and over and over again, stay in your lane. Do the thing that engineers do. Don't do the other thing. Don't be curious about the other thing. Here's what's expected of engineers. And I think this is the kind of engineer that just does not thrive in an environment that is ambiguous, where you're expected to actually do everything. It's sort of the opposite of before. sometimes these engineers really succeed, but like I said, it it takes months of unlearning all all this baggage that they accumulated.
69:11 >> It sounds like the humanities will become increasingly important to develop this kind of judgment and attitude and orientation. >> Have you seen any of that? >> it's possible. you know, we have we have a lot of people on the team that have all sorts of different backgrounds. You know, we have we have philosophers and I'm an I I studied economics. I didn't study computer science. I think we have like an opera singer on the team. So there's, you know, all sorts of different backgrounds.
69:43 >> okay, let's we have time for a few more questions. Let me just look at this list here. I was so engrossed by your answer. there's a question here's okay this this is a very interesting question to what extent do you believe that cognitive offloading might be negatively affecting developers in other words do you worry about clawed code making developers stupider >> >> I think of I think of this book called The Accelerondo >> by Charlie Straws.
70:27 >> By Straws. Yeah. it's just like a >> it's a beautiful book because I think it captures this feeling of liftoff just better than any other book. it it's it's kind of dystopian and I hope the future does not go that way. >> Yeah, me too. >> But it captures the feeling. and at the beginning of this book, there's this guy kind of skateboarding around on this like hoverboard or something and he has his brain essentially in a fanny pack. It's like a brain computer interface and part of his brain is externalized and it's able to do he's he can launch agents. He can do all this stuff just by thinking about it through like a neural link kind of thing.
70:58 and I think this book was written in like two 2006 or two like early 2000. >> I can't remember. And it it wasn't that hard to imagine this kind of future because I think this is actually the place where we've been going for a while, you know, like I used to be able to multiply large numbers in my head, but now now I don't because I open my phone and I use the calculator app. I used to remember all my friends phone numbers, but now you know now I don't except maybe some childhood friends, >> right?
71:24 >> cuz I just have a phone book on, you know, on my phone and as a result I'm able to think about different things. And I don't think one thing is necessarily worse or better. It's not necessarily better to be able to do like wrote math in your head or you know like memorize a bunch of long form poetry. We have like nostalgia for that. But I think because we don't spend our time learning that and doing that we're able to think about higher level things and we're we're able to learn different things and study differently. we wouldn't be able to do that if we just had to kind of memorize everything all the time.
71:54 >> There's just no time for it. So I think this kind of this balance of what we do in our heads versus what the computer does in its head, it's always been this thing where technology changes that balance. and it shifts some responsibilities from us to computers, but I think always it's freed us up to do something that is more human in return. there's a there's a question in here about sort of broadly about competition competition of the sort of u models that anthropic is producing and building products on and like openw weight models and other models produced in China that are competitors.
72:44 so I guess the question is really about, you know, what does that competitive landscape look like from your vantage point? How are you thinking about that? >> Yeah, I mean, I think from my point of view, you know, like I'm on the product side. I think a little bit less about model training day-to-day. I think more about products. >> I think competition makes us build better products. so it's it's really important. we welcome it. we're always inspired by competition.
73:13 for me personally, I actually I'm I have this sort of like weird thing that I do, which is I actually don't use any competing products. and this is because I don't want to be distracted by the ideas that they have because I think it's very appealing when you see a competing product to just copy it. This is the instinct when you're building product. it's just like any product person that builds something in a competitive space, that's always the instinct. It's so easy just to copy a feature. And so personally, I don't use any competing products. So that I have to talk to users. I have to talk to companies. I have to see what everyone needs and and actually solve their problems as opposed to just kind of copying. on the team, there's kind of space for everyone. So there's people that use other products that are inspired by them and you know, everyone in between. And I I think just again competition makes us better. Do you feel that broadly the kind of alignment that we've been talking about in the context of AI safety will be a critical competitive advantage?
74:13 >> I hope it's more than an advantage. I hope that it inspires other labs to follow suit. one example of this is, you know, I mentioned prompt injection earlier. we've now combined essentially model alignment with prompt injection probes. You know, this is kind of monitoring the the neurons of the model with auto mode that that I talked about. And now in cloud code, prompt injection success rate is zero because after multiple years of research and a lot of investment in this, we can no longer demonstrate a successful prompt injection. we've hired external security researchers. We had bounties where we gave out a $20,000 prize to see if researchers in a week can find a prompt injection vulnerability in our models and they could not.
75:01 I think the vulnerabilities probably still exist, but they're really hard to find now. >> Mhm. >> And so the thing that I did actually is I I made a chart of this and I compared our models and the competing models and then I tweeted this >> because I sort of want to embarrass other labs >> into thinking about this. It's like a super important problem. I think this is just one of many tools, but this is kind of one way to think about it. And I think the louder that people are that they care about safety, that they care about prompts injection, they care about alignment, I think this is sort of the it can be this economic force that forces other labs to make better decisions.
75:34 >> Thank you. I'm afraid this is going to have to be the last question. but thank you so much. and it reads, "What would you tell a secondyear computer science student to do? Keep on studying or switch departments?" First off, what I would tell them is just be curious like use all the tools. don't shy away from agents, don't shy away from, you know, quad code. Get to learn to use all the tools so you're familiar with them. the second thing I would tell them is, yeah, probably get a minor in something else.
76:14 >> Okay. well, let's welcome Mark Ekhine back up to the stage for some closing remarks. And let's also thank Boris very much for joining us. >> Yeah, thank you. Thank you, Boris. Thank you, David. A really thoughtful and important discussion. And I really appreciate you sharing your time and your perspective. And I also want to give Vor a thank you cuz I know you've been helping us collect this unique moment in computer history. So, thank you for all that. Let's give one more round of applause.
76:55 We we love all your feedback for tonight. If you go back to the QR code, you can give us feedback on tonight's event. It helps us plan other events. and then we'd also I want to thank Mark and Mary Stevens for supporting tonight's event. but we really do need your help. We have other events. We we have a big museum to run. We have a collection. It's such an important moment in history. We would love for you to join and be part of our community. We keep our membership fee at $15. That's you know, you can buy not that many tokens with that, right? So $15 will get you membership here and be part of our community. We really appreciate it. It will help us support future events, not just on AI. We have two events that I want to share with you.
77:40 October 10th, we have a car day here. It's a Saturday. It's a great day to bring the family, bring your parents, bring your kids. we're not just doing a car day because I love cars. we're doing a car day because they are computers on wheels. and we're have Whimo here. We have Zuks, Tesla, McLaren. We have some simulators, race car simulators. So, it should be a really fun day here. And then on November 5th, we'll have board member Dana Boyd. She's got a new book called Data is Made Not Found. It's on the 2020 census. and DJ Patil, who was the chief data officer for the United States, will be doing the moderation of that. So, it should be a really interesting event. Thank you again, Boris. Thank you, David. Thank you all for coming. Please travel safely for home. And just thank you for supporting the museum. Have a great evening.
78:33 Thank you so much. >>