Vollständiges Transkript anzeigen (7.256 Wörter)
There’s been a lot of misinformation about AI.
This is Andrew. He co-founded Google Brain and Coursera. His machine learning course has reached millions of learners, and he is one of the most influential voices in AI today.
So, a handful of leading AI companies have been very loud voices, fear-mongering around AI, to try to get regulations passed. This drumbeat of fear-based messaging has skewed society's perception to be really negative on AI.
People talk to me about data centers and job loss. Maybe AI could do 30%, 40% of many jobs. And what that means is, well, that 60% that a human does has become even more valuable. What about loss of human control over AI? I think about something else that we can’t control.
I think you're one of the voices in AI who comes with a huge background in machine learning and teaching AI, and also you're a positive voice, because this is something that I've been seeing, especially this summer, how polarized the society has become, especially on social media. When I'm posting about AI and people talk to me about data centers and job loss, why do you think this wave started recently? What do you think the causes are?
There’s been a lot of misinformation about AI, and the root cause of a lot of this is an unfortunate attempt, that started two, three years ago, of I think, PR and regressive capture. It turns out that the most—one of the most valuable things in AI right now is the giant AI models, giant large language models that some companies have trained. But if you spend billions of dollars training a model, it's really inconvenient if someone else trains a model and wants to give it to anyone in the world to use for free. So, a handful of leading AI companies, I think as you know, have been very loud voices, fear-mongering around AI, to try to get regulations passed to create an unfair playing field that favors incumbents, so that we all have to pay a high toll for use of AI while stymying the other teams, be it researchers or other companies that want to just give away open weight or open-source models that anyone can use much cheaper. Unfortunately, fear-mongering works. Um, when you go and say, "AI is like nuclear weapons," which is an analogy that has no basis in fact—what do they have to do even do with each other? Or when you go around and cherry-pick cases of AI making a misstep and making it much bigger than it is, or even spread misinformation about how AI uses data centers uses a lot more water than the actual reality, this drumbeat of fear-based messaging has skewed society's perception to be really negative on AI, which is unfortunate because this is slowing down American adoption in AI. This is making America less competitive, um, and unless we get the truth about AI out there, which is that it's fantastic benefit with some problems, but not nearly to the degree that they've blown up to be, it will, um, hurt individuals.
I'm going to read out some of the problems that people are highlighting: job loss and inequality. What do you think?
The job apocalypse—job apocalypse, this idea that AI will take over 50% of jobs, people will be out of work, writing in the streets. That's just not going to happen. With every wave of technology, including AI, the skills we need to do great work shifts. And so, AI is changing job professions. But, boy, I wish AI were—AI just doesn't work well enough. I know that a handful of businesses want to hype up AI to say, "We have superintelligence," or "We have artificial general intelligence," or whatever, and can do all the stuff that humans do. I wish AI were better. It's just not good enough to make AI do everything a human does. And if you look at the analysis of jobs, economists, uh, like my friend Erik Brynjolfsson at Stanford, and Andy McAfee at MIT, um, economists have analyzed many people's jobs and by breaking it down into individual tasks, and maybe AI could do, you know, 30%, 40% of many jobs, and what that means is, well, that 60% that a human does has become even more valuable because there's a kind of economic complement to the 30-40% that's now cheaper. And so, what will happen is, people that use AI—maybe people that use AI will replace people that don't use AI. But AI is not in a position for the vast majority of jobs to replace people. Of all the different professions, the one that's most affected by AI now is software engineering because AI is actually fantastic at writing code. And what we see is that the number of job openings in software engineering is up, contrary to what, you know, the doom fear-mongerers would say, right? AI is not actually able to replace software engineers, and all the good software engineers I know are busier than ever. Now, the flip side of it is, if someone still write code like it's 2022 before ChatGPT, they're in trouble. They need new skills. Like, don't—don't do stuff that that 30-40% that AI can automate. You've got to stop doing that, let AI do that. But then gain new skills to do the other 60-70% that AI cannot do.
What would your advice be to new graduates? Because I talked to Eric on this podcast, and he was talking about that there is not really a lot of impact on the job market, except for, I think he mentioned, people from 18 to 25 who just graduated. What would be your advice to those people who don't have the expertise maybe to strategize in their job yet? Uh, they can only do manual work that AI can do as well.
So, one real challenge, um, uh, for fresh college grads is that the university system is slow to adapt, and so, um, you know, I love academia. I think we should all support academia and universities. And when AI comes and transforms the way software is written, universities often take like a year or two for the faculty to master skills, then create new courses, get curriculum committee approval, whatever, get the faculty senate to vote. It just takes years, and that speed of change in academia is very poorly matched to the speed of change in AI. So, sadly, many universities are still teaching students to be ready for the jobs of 2022, when we shouldn't even be teaching them for the jobs of 2026, we should be teaching them for the jobs of 2028 and beyond. And what this means is, the job openings are there. Uh, tons of employers I know just can't find enough skill, you know, a people at any level of seniority. But, um, it turns out in my office right now, we have a lot of interns, they're current college students, fresh college grad, works with one high school intern, and they're amazing and productive. But the key is, they're all very AI-native. They all use AI tools to do the things that AI could do, but then also, you know, lean in to doing the things that, uh, humans can do that AI can't for the, for a long time. So there's plenty of work for people to do. But so my advice to fresh college grads or people currently in college is, um, by all means work hard in classes, you know, get good grades, learn from the instructors, but to the extent that there are additional skills that the university has not yet adapted to teaching, then find other ways to learn online, be it from Coursera or DeepLearning.AI or Udemy or other places, where you can gain the more cutting-edge skills, especially AI skills that, um, universities have not yet worked into the curricula.
Quick pause, because what Andrew just said about workflows connects to something that HubSpot just put out for free. The thing is, the real opportunity right now is not just in the models themselves, but in what you build on top of them. How you turn AI from a chat window into workflows that actually do work for you? And the good news is, you don't have to start by building a full agent. You can start much simpler with better prompts. HubSpot just released the Advanced ChatGPT Prompt Engineering Playbook, and the idea is very simple. In seven days, it helps you move from getting generic AI answers to building prompts that give you much more consistent and useful options, because we've all had this moment: you type in one vague sentence, get something mediocre back, and then spend the next 20 minutes fixing it yourself. This playbook is basically built for that exact problem, and it walks you through a single progression. First, you learn how to structure prompts with things like: role, context, format, and parameters. Then, it gets more advanced—few-shot examples, chain-of-thought reasoning, more precise prompt structures, and ways to make outputs more reliable. What I like is that it doesn't stop at individual prompts; it also shows you how to build reusable AI personas, modular prompt components, and eventually your own signature prompt engineering system. So, instead of starting from scratch every time, you're building a repeatable way to work with AI, whether you're creating content, analyzing data, or making business decisions. The playbook is free, link in the description. Thanks to HubSpot for sponsoring this video. And now, back to our conversation with Andrew.
I want to say one other thing. Um, it turns out one—if I look at the skill map changes, one of the most important changes is, um, it's so much easier to build with AI than before. When something becomes much easier, a lot more people should do it. And so now, not only should professional software engineers build software with AI, it's becoming much easier for everyone to build with AI. And people that embrace that and do so will be more productive and will accomplish more and I think have more fun than the ones that don't. And AI lets you build really fast. So, for people that are not just software engineers, but, you know, marketers, recruiters, uh, HR professionals, operations specialists, I think if they learn to build with AI, uh, they will just do much more, whatever the job role is.
How do you, by the way, measure, uh, the increase in productivity, uh, when you deploy AI? Do you have, like, a KPI in your company?
I wish there were a simple answer. I find that the business outcome of AI is more a function of the business than a function of the AI. For for some, it may be an increase, you know, uh, customer growth and retention, or maybe a faster to serve customers, or it would be an increase accuracy in some tasks. So, the KPI tend to be related to the business rather than the AI.
Um, so you can't like directly measure just AI. Because it's it's an interesting thing to do because we've been deploying AI actively in my company and I think from me as a media company, it's probably the amount of views, the output. Uh, it's just interesting, how, yeah, it's just interesting how different people measure, even revenue, like if you're becoming more effective, uh, with, uh, how you make money.
Actually, say more how you're using AI in your offices.
Oh my god. Uh, I have the—so, first of all, we have Claude for all of us and we have certain projects for every social media that we're on. So for example, for this podcast, we have a project that's called "Guests," and it knows all the analytics from previous guests and it has certain criteria on which we rank every single person who comes to the podcast, whether he's he or she is cited, whether they have a certain opinion on AI, whether they've been active with AI in their company or if they're a recent founder in AI, so it gives them different weights, and it comes up with a grade based out of 40, 40 meaning tier 1, 30 meaning tier 3, etc. And then we have another one that analyzes every single podcast and gives me tips on how to ask questions. Oh wow. Same for Instagram, same for LinkedIn. It has my tone of voice, personal dossier, my business strategy, so it whenever it writes something, it knows all the facts about me, how I sound. Every social media is run by a person, so a person makes a strategic call. And by the way, if you can give me feedback on this, if I can improve. So what I'm working on right now is closing the loop because sometimes they send me a text, and I'm like, "Oh, we need to change this, this, and that." But that happens in a chat in Telegram, and we have this feedback. We have a bot that scans all of our chats. But I really want AI to be able to learn continuously from this feedback to just know my taste better.
There’s one thing I see a lot in AI, which is, um, it turns out, for AI’s data scientists, or AI’s brainstorming partner, it often comes up with, you know, one or two good ideas, two or three mediocre ones, and like, you know, four atrocious ones. And sometimes you wonder, "How could my AI have thought, you know, like that could even be a plausible idea?" And to me, this relates to the job apocalypse point of view, which is that, for a long time, humans, you, me, everyone watching this, we have a significant context advantage over AI, which is that you know something that's incredibly obvious to you that, you know, that was an awful idea, but the AI did not. And it turns out that one of the reasons why AI will not replace all jobs, or whatever, or a large percentage of jobs anytime soon, is because humans have a massive context advantage compared to AI. We know so much that, you know, from years of experience, that we talk to customer, we saw the funny facial expression that tell us, "Ah, they don't like this," or we talk to business, or, you know, our manager said, "Blah blah, I really care about this." And so it turns out that almost all humans, what, maybe all humans, just know a lot of stuff that the plumbing does not exist, and I don't think exists for the foreseeable future, for AI to get. I know sometimes people talk about the importance of human judgment or human taste, and some to want to, "All right, what is taste?" This is this fuzzy thing. But to me, the technical thing that underlies why humans have better judgment and better taste than AI, is this context advantage. And because this is a long-term advantage, like no one's going to solve this, you know, in a few years. This is why we just need a lot more humans with that judgment and taste to keep on complementing the AI.
And doesn't this make education even more important? Because education gives us context. Because it's another, another thing I'm hearing about AI, like, "You won't need education," because all the information is at your fingertips, you just ask ChatGPT. But when you say context and taste, for me, that's years of acquiring knowledge and learning from the best and seeing how they perform versus just asking a chat.
I'm going to say something that may be controversial. I don't have said this publicly, but I think it's true, which is, frankly, AI models are terrible for learning. Um, I know people think AI is—AI is wonderful at getting things done. Use it all the time, love it. But all the data that's coming out is that when, say, college students use AI, we know this, it's just a study, it's now back to this wall, so we also have numbers, but the data is very clear. Students score higher on homeworks when they use AI. Yay, higher homework scores, but retention, their long-term performance is much worse because they AI do the work for them. More and more studies are coming out to back this up now, that I think people think, "Oh, it turns out, you know, I think Wikipedia is a wonderful tool, has tons of facts. A web search is a wonderful tool, has tons of facts." But it turns out that when you ask AI to do work for you, you're cognitive offloading to AI, which is great because that's how society moves forward and gets work done, but human retention is much worse. It's just so clear that LLMs, as they are most commonly used, are terrible for learning. I'm not saying there's no way to use it in a way that's good for learning. I think there are ways to use it that are good for learning, but even for myself, there's so many things I asked the AI model over the last, you know, six months or whatever, like, I don't know, building some project, how does this front-end, back-end, component work, whatever, give me the answer, get the job done, it was fantastic. But six months later, I don't remember the answer. When I need to redo that front-end, back-end component, I ask the AI again. So, the data is very clear. We should stop thinking of AI as helpful for learning, at least the vast majority of ways that the vast majority of people are using AI models today, it's absolutely terrible for learning.
But you're building a company helping solve that, right? Because the one-to-one tutoring with AI is that where you just announced with the 100 million investment from Coursera.
Yes. So, I’m excited about leading a new organization called LearnVector that is focused on, um, building new learning experiences that is much more one-to-one than one-to-many. So, you know, 15 years ago, I—I was privileged to participate in the online courses movement that I think changed the way a lot of people learn. But that was and still remains largely a one-to-many experience where everyone, you know, kind of watches the same video, which is actually okay, that actually works well. But the technology now exists to create much more personalized, customized, one-to-one experiences. And so, our team is working hard on that. I think we’ll have a—a lot more to show by early next year.
When think about human skill development, I—I feel like because AI has so heavily impacted software engineering, um, what we see happening in the job market for software engineering is a harbinger or a fore-runner of what we'll see in other disciplines as well. And in software engineering, um, people need to learn new skills. But when they do, they're thriving and creating more value and, frankly, getting raises and doing even more exciting projects. And what I'm seeing the early signs of in other disciplines as well. For example, in software engineering, you know, most developers, like front-end, back-end developers, have now become full-stack developers because of AI, help you could take on broader scope. I'm seeing early signs of this in other disciplines as well, where, for example, someone that in marketing that did marketing coordination, uh, coordinating marketing campaigns, with AI help, they can now become more of a full-cycle marketing, take on a broader scope. And I'm seeing, you know, frankly, sources in recruiting become more full-cycle, do end-to-end recruiting. So, now, the good news and bad news is for people to step up to these broader roles, you do need to learn AI skills, but also, it's not just learning AI, you also need to learn these other skills, uh, like how to do the other parts of marketing, of recruiting, or software engineering or AI engineering. So I think this actually creates a heavy need, a big need, for people to gain new skills. But when they do, which is both AI skills, but also disciplinary skills, then they can do much more, hopefully have more fun, work on more exciting projects, hopefully get paid more as well. And one reason I kind of worry about the fear-mongering is, um, I got an email from someone that was about to enter college, and, you know, he emailed me saying, "Hey Andrew, taking online courses, but I'm really struggling what I should major in college because in four years, won't AI do all this and everything I learn will be obsolete?" And the answer is, no, of course it won't all be obsolete. But when we keep on pushing these fear messages, uh, we make people wonder if they will even be relevant, and it makes people not lean in to gain these skills that will put them in a much better position. So, I see very clearly that these fear-mongering messages are distorting how many people, including, you know, high school students, college students, fresh grads, think about the economy. And frankly, making people give up is one of the worst things we could be doing. In this era, when people that lean in will thrive.
Andrew has taught over 8 million people AI. He started teaching machine learning online back in 2011, years before the current AI boom. Now, one of the companies he's building is focused on AI agents. From the way that it sounds, it can still feel way too technical. So I put together a step-by-step guide to building your first AI agent with no coding required. It walks you through what to automate, how to set it up, and how to make it actually useful. It is in my newsletter this week. The newsletter is called Future Proof, it's free, link is in the description.
What would you reply back to that email that somebody sent to you? What would you say is the best major to study now? To thrive in AI era, do you think it's like going deep in to niche or just broader computer science so that you can acquire AI skills really fast?
You know, I don't know what's the best major. There are an awful lot of great majors. It's it's like, um, I kind of feel like what's the best job in the world? It's like what's the best major in the world. It's something that you love, right? Yeah. My daughter wants to be an astronaut. I don't know if she can major in becoming an astronaut. I have to think about that. When she gets older, she may change her mind. I see so many opportunities, um, across, across so many job roles. It all seems very exciting to me. But do learn AI, do learn to build with AI. The other thing that my team is working on the AI engineering skills map to try to map out, you know, the most important skills for AI engineering. One thing that I felt intuitively, but I was surprised to see it show up in the data was that a lot more job descriptions seem to be saying they want people that demonstrate a very high sense of agency. Because it turns out, with AI, there a lot more opportunities for individuals to spot problems and go build something or do something that goes solve it. So, I think we're really evolving, what we've long been evolving, but we're accelerating past the era where people sit around and wait for their boss to tell them what to do.
This is what I've been feeling a lot, especially when we started doing remote work. I want people to be entrepreneurs within their niche. Like if you're helping me with LinkedIn, you're an entrepreneur there, you can hire more contractors, you can deploy different tools, you make the strategic decision whether this topic is good or not, shall we proceed with it. I really think, and tell me if you agree with me, we're moving into that job market where everyone is kind of independent in their workplace.
I think people will have much more autonomy and creativity. So I agree with that. And I even go one step further, which is, I talk to a lot of people is, the engineers and others in large companies, that tell me that their manager tells them to stay in their swim lane. They'll say, "Oh, I have this creative idea, but their manager is, 'No, I need to focus on this one thing.'" Frankly, often because their manager's career depends on it. But I feel like the number of opportunities for people to spot things outside their swim lane, um, and then in a responsible way explore how to get it done, that feels very exciting to me. And I think that in the future, the businesses that set up a culture that encourage people to learn AI, build fast, responsibly, um, talk to customers, will drive a lot more value than the more hierarchical silo organizations.
Yeah, it—it starts with hiring the right people and then nurturing this in your organization. When you say, learn how to use AI and become professional with AI, can you give me some benchmarks? Like of a person who's, like, say, marketer, knowledge worker, advanced with AI? What are you looking for when you're interviewing this person?
I'm pretty sure my team is ahead of the curve. All of my marketers know how to code. So, as part of how I interview marketers, we ask them what they've built. And, uh, if they have not built any software.
If it's a dashboard, is it good or bad? Like, is it too basic or?
It's okay. A dashboard's, again, my team's pretty, you know, somewhat ahead of the curve. But, um, all of my marketers have built much more sophisticated things than dashboards. Oh, I—I feel like, um, I don't know, the other day, uh, one of our, someone on the marketing team was talking about the tools that he had built to, uh, when he's considering writing an article on something, it will, um, crawl the web, find related work, has a custom desktop app, actually built a desktop app that runs on his Mac, to, um, highlight related articles for him. Then he can chat with the whole system, navigate, you know, the thing he's writing as well as the related work, and he has a large dashboard for crawling the internet to highlight to him exciting things that are popping up. Um, now, even on my team, I think that marketer is ahead of the curve.
But that's—that's great to hear. Any other interesting use cases, uh, that will inspire people to build something similar?
Let's see, maybe a, uh, my finance team uses AI extensively. So I think a—a, uh, one of my CFOs, uh, realized that, you know, her team was spending hours every week clicking through documents, open this, copy-paste this number here. And so, um, she started building, uh, automation scripts that runs on a routine that, um, automatically opens files, checks what's in there, checks the consistency, highlights to her team if there's something, uh, if there's something they need to be paying attention to, if a new document that showed up. So, I find that rather than waiting around for an engineer to do the work for them, the team's ability to, to kind of a, not just build dashboards, but build, kind of a, data management infrastructures that can adjust data, alert them if something is happening. Yeah. I think, uh, my finance and marketing team are doing that. Oh, my recruiting team, um, well, we actually have recruiting engineers, which are really professional engineers that sit in a recruiting team, that are building very sophisticated tools for recruiting. And this is actually the other trend. I think marketers, recruiters, HR professionals, ops people, should all learn AI. But the other thing is when you take an engineer and embed them in these teams, then that further accelerates what you can do.
We do the same. We—we start with something basic, build it ourselves, then we hit the wall, and an engineer comes in. We build it further.
Frankly, when you look at not just software engineers, but recruiting engineers, marketing engineers, HR engineers, I think there's so much valuable engineering work that can now be done. I'm just, you know, not worried about running out of—frankly, all my friends are so busy, we think, "Boy, how can we run out of engineering jobs?"
Yeah, yeah. There are so many cool ideas you can experiment on. But you touched upon something that is actually one of the fears. When we talk about, like, financial information, how much you're giving to AI. So I gave my, Perplexity, permission to scan my fidelity account, so it can track my portfolio, tell me when to rebalance. It doesn't do anything on my behalf, but it has access. Do you think there is any problem with that?
This is complicated. I think AI and privacy is a complex area. Um, and it depends a lot on the company that you're sharing your data with. So, for example, I trust all the hyperscalers to really 100%, you know, follow their terms of service and to do what they say. My personal opinion—not giving legal business advice—but I’d be shocked if, you know, the largest hyperscalers publish the terms of service, or some privacy notice, and if they breach that. That, because that would be not the culture, be so damaging of the long-term business model. Now, that's on the largest hyperscaler side. If you look at AI company side, there's been, you know, at least one company that I won't name, that seems to occasionally change the terms of service and you're using it, you go to the website, so you pop up, "Hey, we changed the terms of service to retain your data, or train your data." And if you're not paying attention and click the wrong button, then they suddenly gave themselves permission to access your data in a way that I'm not that comfortable with. I feel like I handle, you know, some sensitive information, so then I tend to be very careful with the businesses that I just don't feel their culture and their DNA, and frankly, the long-term business model, is as tied to protecting individual user privacy than the hyperscalers. And I see businesses, you know, get this as well. For example, one of my teams at AI Aspire, we work with very large corporations, including banks, with incredibly sensitive financial data. And, as you can imagine, AI Aspire and and our and our clients do not willingly share, you know, very sensitive, I often material non-public information, right, and NPI, uh, with Frontier, with with a Frontier labs without really careful thinking about the guardrails and privacy. So, that thing is complicated.
So, trusting hyperscaler, but also, um, another thing that you can do, you can download an open-source model and just run it on your computer, and then it just stays on your computer, right?
Yes. I think, yes. It it turns out a lot of banks will actually run, um, the things in, uh, you know, like a virtual private cloud, or on-prem. So, they never even leaves their control. But I think for individuals, it’s, you know, for the really sensitive things, um, uh, I sometimes run a local model, and it's been interesting with the open-weight models. So, the latest open-weight models are approaching frontier capability and that are, you know, actually small enough, they're actually really good models now, they can run on your, the one from Meta, right? The recent one. Oh, yes, Meta Muse is a good model, and I'm thinking also the latest version of Qwen is also very good. But I think frankly, these models change every other week. So I think the best practice is to not get stuck on one, but to keep on trying new models.
So basically, when there's a situation that you don't trust anyone, you run a local model, and this is how you keep your data safe.
I do trust the hyperscalers, but sometimes for, you know, literally NPI, material non-public information, that I won't even send to the cloud, so that, uh, I’ll either do it manually without AI help, or if I really need to use AI, then, you know, really carefully, only use a local model. Yeah.
Interesting. Okay, this is this is an interesting one. Okay. What about loss of human control over AI? Because I've talked to Yoshua Bengio, who is very um negative when it comes to open open free AI without any regulation, and he painted me some very scary pictures of AI taking over control because we basically the the whole scenario is we can't control something that's smarter than us, and if AI gets smarter and smarter, where where do we end up? What do you think about that?
I think about something else that we can’t control, which is airplanes. No one can build an airplane that you can fly perfectly. Winds will buffet it around. And incantantly, in the early days of developing airplanes, some airplanes crashed, and people died, and it was tragic and awful. But, through the early lessons learned, we then learned to control airplanes better and better so that today, you know, we can mostly get in an airplane and not fear too much for our lives. And it's really like that too with AI. No one can perfectly control AI because it generates tokens or outputs that are a little bit random, so we don't really know what exactly they'll do. But as we run them, and you know, there's been a small number of mishaps, which is unfortunate, and some number of mishaps have done some role damage, but the way we engineer almost any system, from an airplane to electric circuits to now AI, is carefully grow their capabilities so that we can have a controlled environment in which to measure what's wrong and then to shape it to make sure we can control it well enough that it behaves responsibly and safely. And to this day, we can't perfectly control any airplane, and we will never perfectly control AI either. But I think that we are certainly controlling them well enough that this loss of control doesn't feel like science. It feels like science fiction.
Yeah, yeah. What about deepfakes?
Deepfakes are a problem. Well, one of the most disgusting things I've ever seen or heard of is non-consensual intimate deepfakes. And agree, uh, I'm really glad that, you know, US Congress has been moving to, right, let's pass laws, get rid of that, penalties for that. I'm just—I think there's some really problematic uses of AI that we should outlaw, heavily penalize, let's just get rid of that.
What do you think about children and social connection when it comes to AI with kids using more of AI? Because we've seen social media how, you know, there are people who are dumb scrolling all day, and my daughter who is five years old now, whenever I don't have an answer, she's like, "Ask ChatGPT." And like, who's that person? I like, I don't know. Ask ChatGPT. And like, she thinks ChatGPT knows everything. What would you say about, you know, kids' future with AI?
First, I think kids have a bright future. There's such an exciting time to be a child, to grow up in this environment with tools that none of us ever have had before. At the same time, we've seen that social media, um, I think social media has probably been blamed a bit more than it deserves, but it does deserve blame, uh, has kind of not been great for kids. I actually worry a lot about... It's a wonderful tool, but AI damaging learning is something I worry a lot about. So it turns out, um, I have a five-year-old and a seven-year-old. When I teach them math, they're still young enough that I can, basically, you know, not let them use the calculator. Can say, "How do you multiply these numbers?" And I don't give them a calculator and practice that with them. But as they're a little bit older, I worry a lot about students using cognitive offloading to AI in a way that damages the long-term learning and retention. Um, but then at the same time, oh, I actually built an app. I did not like any of the, you know, free online learning to type, types of things. So, I actually built my own to have my daughter learn to type, and I'm hoping that she's actually getting pretty decent now for a seven-year-old. Oh, yeah, she can actually type all the lowercase letters. She's still a bit, you know, not, her shift uppercase letter is a little bit not quite there. But I think that this, um, unlocks, you know, responsible adult-supervised use of online tools, and I think it is really tricky, you know. I think um, adult-supervised use of digital tools seems a great thing for kids. But too many adults don't have time to supervise the use of the tools, and then the incentives of, say, social media, right, to do funny things.
Yeah, it has to be the right incentive when it comes to AI. Okay, you mentioned—we talked about the fears, we talked about how you can improve your work with AI. Can you name some of the biggest opportunities in AI in 2026 for people who want to build?
For an individual that wants to build, I don't think it's one size fits all. But because the cost of building has plummeted, um, I encourage people to learn AI, build fast, and talk to customers. I find that I’m building things, I don't know, every week, every weekend, because I or someone on our team has some problem and I have some idea for building some AI thing to automate it. Last weekend, I had, really as using a frontier model to analyze a lot of our key business metrics because I didn't have time to do it myself, but it was kind of, measuring, you know, DeepLearning.AI's key business metrics and I didn't have time to go find a data scientist to go work with me on it, so I just used a variety of frontier models, being really careful on their uh, data retention policies, I did not use models with data retention policies that I don't like, in order to analyze data. But and then I find that, um, what's happened with AI is the cost of building has plummeted and so the challenge is shifting to deciding what to build, which I was calling, which I've been calling the product management bottleneck. And so people, your founders, engineers, product managers, that can talk to customers, get a sense for the taste of judgment on what to build, and build with AI and iterate quickly, I think there's just a ton of exciting things to be building. And you've been starting so many companies. You're like, when I looked at your portfolio, do you think for beginners when you said you built something during the weekend, how do you decide what to focus on? Or you can pursue multiple ideas because of AI now and you can just be, you know, playing in different companies at the same time?
It turns out building a company is still really, really hard. And so there's a lot to be said for single-threaded leadership or someone that's fully focused on just one thing. I find that, you know, over a weekend, I can often build an own wrapper, build a simple application, but I wish it was that easy to build a large company. Um, I find that building something meaningful often takes either real technical depth uh and or deep customer insight and integration with customers. And yes, we can now, you know, use AI to code something in a few hours, but that's a small piece of the puzzle. So spend time understanding the technical complexity and building the really complex software, that takes us like months, you know, maybe years, or having that deep customer insight to decide what to build, that also just takes a lot talking to people, reading facial expressions, surveys, doing that over and over until we figure what to build. Um, and so, I think there's sometimes a lot of value to sampling widely, but then having that focus for an individual to go really deep in the couple sectors, that that still seems important for building a business.
My last question, I know it's we don't have much time, but I wanted to ask you about AGI, just because people use this word so much and some people say—I think Jensen Huang said we already reached AGI. You said it's decades away. What's the one criteria when you're going to say we reach AGI?
So, different people say we reach AGI at different times because of different definitions for AGI. The definition I'm most familiar with is AI that could do any intellectual task that a human can. But so the human brain can take, say, 5 years to study and do a PhD thesis, uh, or and so can AI write a PhD thesis? Or a human can learn to drive a truck through a dense rainforest with, you know, tens of minutes of practice, so when can AI do that? To drive a new environment with tens of minutes of practice. It feels like there's a long list of these things that AI cannot do, uh, for what feels to me decades, I hope it's only decades, maybe it will turn out to be longer. So that's what I think for that definition of AI or AGI. AGI is still very far away. But, um, it turns out because of economic incentives, I think OpenAI and Microsoft had an agreement that's actually been renegotiated now, so that's gone away. But OpenAI had the economic incentive to try to declare reaching AGI earlier, uh, and so it turns out that if you come up with other definitions of AI, depending on how far you low the bar, then you could totally have reached AGI, you know, already or even 30 years ago, depending on how you want to define it.
Yeah, true. Andrew, thank you so much for this positive conversation, very applicable. I like when, um, you watch something and then you go and you measure yourself against what people are doing with AI, look at your process and, uh, maybe expand it. So, thank you so much for showing what your team is doing and thank you for your insights.
Yeah, I think given the huge benefits of AI to come, I hope whoever is watching this is motivated to really go learn AI, apply it, um, and and even to go build some things.
Video
