Reid riffs on Tobi’s memo, AI and play, and the tweet that cost trillions

0:28I'm Reid Hoffman. Reid, so lovely to be here with you. So recently, Toby Lutka, the CEO of Shopify, put out a memo that was covered by everyone because it had to do with internal employees at Shopify and what they were expected to do with AI. It talked to them about how if you're going to request more resources for your team, you actually better check if AI can do the job better, faster, and you actually don't need that additional headcount. It also said, like, everyone here should be expected to be using AI every day. And he said, I'm the CEO. I'm no different. We want to grow by 20, 30, 40 % of year.

1:06Every employee needs to grow as well. I think some people were shocked by this memo. Other people found it reasonable. What did you think about the contents of the memo and also Toby sort of putting out this bold statement for the industry? You know, I also found Toby's memo to be exactly right. I thought, you know, it's kind of the kind of leadership that, you know, Toby does. And then also thinking about kind of his classic technologist, because he's obviously an engineer, is how we use tools, you know, AI as amplification intelligence, how it is that we get, you know, super agency through doing this.

1:40And his memo, I think, is exactly the kind of thing that I think everybody, not just technology companies, should be doing. Every single CEO of anything from a, you know, call it, you know, five to seven person company to a tens of thousands of people company should look at that and say, what's my version of how I should do that and how I should integrate it? And Toby obviously has given it enough thought to kind of say, look, here are some, you know, key, you know, kind of checkpoints that work within companies, which is you ask for more resources, make sure that you have, if you're going to ask more resources, how you're asking for more resources in a context of, and here's how I'm already using AI, and here are the reasons I need more resources given how I'm using AI, either AI's lacks or the AI opportunities from doing it.

2:34So one of the things that I've been telling, you know, kind of my portfolio companies is to actually have kind of weekly, monthly check-ins where everyone has to bring a, And here is the new thing I've learned about how to use AI to help me do my job, help us do our job, help us perform better in our mission as a company for doing that. Because the answer is, if you actually haven't found something that was useful to you, it's useful to your group, it's useful to your company, you haven't tried hard enough. I think Toby's memo is the kind of thing that, in fact, CEOs and all group leaders should be looking at and saying, great, how do I build on that?

3:20Thank you for the open source kind of management technique. What are the things that I should do specifically for our group, for our company, for our mission, for our culture? What is our version of that? And then start iterating in the same way. I mean, I have to admit, last week on LinkedIn, I saw a marketing agency. And they said on LinkedIn, you know, we promise our clients that you will never get an image that was started in an image generator that was using AI. We promise you you're never going to get a tagline from us where we use ChatGPT to create it. And I literally had to look at the posting date because I thought it was April Fool's joke, and it wasn't.

4:01And I get the nervousness and the scared about your job and about the future, but I just sort of couldn't imagine that this marketing agency was essentially doing the exact opposite of Shopify and sort of banning AI in their workplace. I'm sure it befuddles you as much as it befuddles me. Well, I mean, I think generally speaking, that's the similar idiocy in the education space saying our students shouldn't use ChatGPT because the whole answer is you're preparing them for the future. You're preparing them for being citizens, for being workers, for being, you know, you know, people who are navigating life.

4:37And here is this fundamental tool. It's kind of like saying, hey, none of our people can use anything that uses electricity. And that's how they learn. They have to use pencils and papers and no electricity whatsoever in anything. You're like, Well, that was idiotic. Well, it's similar to the ChatGPT. And so I think that marketing agency, the question is really, when is it going to have to shift or it's probably going to die or become very esoteric boutique? Right. Absolutely. If you want to be the most boutique agency, perhaps that's the way to go. So another concern people have with AI, though, is misinformation, disinformation, all of this synthetic media that was created.

5:16And actually last week, and this wasn't created by AI, this was a tweet that sent the market, made an$8 trillion worth of market volatility because someone tweeted that the tariffs were off when they in fact were not. And so if a single tweet can move the market by$8 trillion, like what does this mean for the future when disinformation, misinformation is increasing and perhaps with algorithmic trading and AI able to do this at sort of greater quantities and greater speeds, like how do we protect against that for the future? There's a combination of a free market response, which I think is partially correct, and a societal response, which is also partially correct.

5:59And so that's the balance that makes this challenging. So the free market response is simply say, well, if people who are doing trading are going to be idiots and not track false posts and so forth, they're going to lose money and eventually they will be disempowered. And so what you principally need to do is to just make sure that there is validated sources of information, you know, kind of are the anchors and then increase that validation, you know, accuracy, availability, and then allow the market to sort it out. And that's a partial answer. And my principal thought there is, like, we should not be trying to restrict technology as much as we should be trying to shape technology.

6:44Because the question isn't like, let's not have algorithmic training. And it's like, OK, that's kind of foolish. It's let's have algorithmic trading work in the following way, generating the following reports, making sure it's involving the following kinds of data is only deployable by entities that have a method by which they participate in the market in a way that is healthy for not creating crazy volatility swings that damage society. It's a little bit like similar to saying, hey, you know, car manufacturers don't want to manufacture seatbelts. Drivers don't want to wear seatbelts. But actually, in fact, because the cost to the society and the health care system and everything else is so high, like you would say, hey, it's a free market.

7:30You should decide whether or not you're going to take the risk and they're going to die. It's like, no, no, no. Actually, in fact, there's so many injuries and so many costs here. And the cost of enforcing you to wear a seatbelt is very low. So let's do that. And like what are the seatbelt parallels for making the overall system work is, I think, an ongoing and kind of thoughtful thing that banks and regulators and intellectuals and economists should think about. What are those minimal kind of, as it were, shaping technology or technology ads that keep the cost of transactions down and the cost of not having an overly centralized system and the benefits of all the free market and broad network working while navigating the fact that we kind of live in a more volatile space now?

8:31On a lighter note, if there's any parents out there who are navigating this, I just read the book Lemoncello's Library with my nine and seven-year-olds. And a main plot point is a fake Wikipedia post that leads to ruining someone's reputation and the kids who don't believe it. So anyway, try that out if you're looking to teach your kids about misinformation on the internet. But actually moving on to another thing that people think is childlike in play, one of the fun things about our conversation with Demis Asabas last week was we talked about games. And it was so clear that Demis grew up playing chess and games were so important to him, both in terms of his scientific research, but also in the progression of AI, whether it was, you know, AlphaGo or, you know, the famous IBM Watson chess competition.

9:14And so when you think about the future as AI is more enmeshed in our daily lives, like, will that give humans the opportunity to play more? Are we going to be playing with AI? Are we going to be interacting it solo with teams as a game? Like, how do you see that connection between games and sort of our AI future? There's a fun book, which, you know, Demis also knows about homo ludens, which is like, we're not just sapiens, we're game players. Obviously, you know, I have this version of homo techne because I think part of games is his technologies and the technologies that enable different kinds of gameplay as part of it.

9:51But games is a way we think. And as you know, you know, I tend to approach like most of my strategic thinking through the lens of games. So it's like, it's like with a startup, what's your theory of the game? With a project, what's your theory of the game? With creating a book super agency, what's your theory of the game? And so because game playing brings tactics and strategies and transformation, like large language transformers together and also has a notion of increasing learning and competence because how well are you playing the game what are the conceptual tools you're bringing to it etc so so games is a way that we operate across you know kind of call it intelligent experience like almost like is species x intelligence how do they play games is actually in fact you know kind of directly correlated to that.

10:41It's one of the reasons why we know that other kinds of mammals and other things have intelligence because we see dolphins playing games. We see chimps playing games. We play games with our dogs and we play games with our cats and kind of that initiating gameplay and everything else is part of how that tends to operate. We don't just play games solo. We don't just play solitaire. We don't just play games one-on-one. We play games as teams, you know, sports games and all else. And that's part of how you model what companies go. And when it gets to this kind of super agency future of saying, well, how it is that we're deploying?

11:17It's like, well, when I deploy now in work, and this is kind of the Toby Lutka memo, it's like I should deploy with agents. I should deploy with these tools. And by the way, we as teams should deploy with these tools. We as companies should deploy with these tools. We as individual scientists, as groups of scientists, should deploy with these tools. And that's the kind of the pattern that we're on. And that's the model of games is a good way for us thinking about it. But it's also a good way for thinking about like, how do we construct these devices? And also, how do we interact with them? You know, part of the original, like the very first genius moment that, you know, Demis and Shane and Mustafa brought to Scalable AI was realizing, here is a way you can apply scalable compute and learning systems to creating amazing cognitive.

12:12As opposed to like, we program the AI, the AI learns. And it learns at scale, because you can use self-play as a way of doing it. And that was actually part of like, seeing this genius moment by them was part of what got me back into AI from my, you know, kind of my undergraduate days where I concluded that the mindsets and programming AI would actually not work. You know, I hadn't gone to, well, what are the scalable compute learnable systems? Because back then, by the way, like a single computer was super expensive, let alone like, you know, how do you create a server farm of 100 ,000 kind of working in concert?

12:57and all the rest. And by the way, the computers back then were less powerful than the smartphone that's in your pocket.

13:11Possible is produced by Wonder Media Network. It's hosted by Aria Finger and me, Reid Hoffman. Our showrunner is Sean Young. Possible is produced by Katie Sanders, Edie Allard, Sarah Schleid, Vanessa Handy, Aaliyah Yates, Paloma Moreno-Gimenez. and Malia Agudelo. Jenny Kaplan is our executive producer and editor. Special thanks to Surya Yalamanchili, Sayida Sepieva, Thanasi Dilos, Ian Alice, Greg Viato, Parth Patel, and Ben Relis.