AI Will Cause Jobs to Evolve, Will You Be Able to Evolve With It?│Justin Wolfers

Terrific, and thank you. It was very kind of you. On a topic like this, the future of our labor market, it's wonderful to see so many young people here because it's your labor market that I'm discussing. Those of us of my age are just hoping to get through the rest of our careers without computers and robots taking our jobs. So, here is the roadmap for today.

We want to talk about the economics of AI. So, I'm going to start with the microeconomics of AI and ask how it's going to change our work and whose work it will change. Then, we'll move to the macroeconomics and ask what sort of big-picture effect it will have on things like how much money we make and how much stuff we produce. Finally, we'll turn to political economy.

This is a polite term for politics, which is going to upend an enormous amount within our labor market, and it's going to create distribution conflicts. It's important to understand where and how those conflicts arise. I also coined a new term: if we're going to transition from microeconomics to macroeconomics, I want to discuss how this affects you. So, I’m going to call that "me-conomics."

How can you adapt to the changes that are coming in your career? Let's start with the microeconomics of AI. The first question is, how is this changing the workplace? I want to begin with three business case studies. The first of these is a story of AI.

What they did was recruit professional freelance coders. They assigned each of them a very simple task: to implement an HTTP server in JavaScript. In this case, the group that received tails was told to just go ahead and code it the way they normally would. Therefore, they utilized the same types of internet resources that anyone in this room who codes would typically use.

The other group was given access to an AI assistant called GitHub Copilot. At the time, it hadn't yet been released to the public. What you see here is that among the control group, it took a little over two hours to complete this task. However, among the treatment group, who had only one minute of training in using this new tool, they were able to complete this fairly straightforward coding task 56% faster, in just 71 minutes. This suggests there may be significant gains for at least some people from using AI.

Let's take another case study. This case study involved a group of office workers and aimed to enhance their professional writing skills—skills that many of us in this room typically possess. Those of us with white-collar jobs often need to write press releases, short memos, or sensitive emails to address performance issues. In this study, the control group was instructed to continue with their regular work, whatever that entailed. Meanwhile, the treatment group was informed before the task that they needed to sign up for ChatGPT and that they were encouraged to use it while performing their assignments.

It's a fairly straightforward task. The people in this experiment were individuals like you and me. They were white-collar workers and college-educated professionals. The average time among the control group to complete this straightforward writing task was 27 minutes. However, among those given access to ChatGPT, it was just 16 minutes.

So it was 40% faster. In addition, the quality of their work was, in fact, somewhat better. The third business case study I want to present comes from a Fortune 500 software firm. This is not an experiment in the lab; this involves real people doing their jobs every day.

This is a software firm with a help desk. While it is typically referred to as a call center, customers do not make phone calls; instead, they engage in online chats. In this study, the control group was instructed to continue their usual work when a customer sought technical support. Meanwhile, the treatment group was given access to a new chat assistant.

Basically, they would continue doing their work, and the computer would monitor what they were doing and offer them suggestions for how they might respond if they wanted to. The person was still in charge of their own work. They could reject the suggestion, follow their usual approach, or, if they thought it was a good suggestion, they might choose to accept it. For instance, the computer might say, "I'm sorry that you're struggling with that problem."

Let's try to troubleshoot it. They could say that instead. What we see here is that they were able to close customer support tickets 14% faster. Hopefully, these three case studies, which are real experiments, show that this new treatment, this new way of doing work, is remarkably more efficient. What I've shown you is that people tend to do more work.

What I haven't shown you is that there is also work and findings in all of this literature indicating that they are producing higher quality work. One of the really interesting aspects of this study on customer support staff is that, you know, when you contact customer service and receive poor service, you might say, "May I speak to your supervisor?" It turns out that the people in the treatment group using the AI assistant were less likely to request a supervisor. This suggests that not only were they handling more customers, but they were also providing better service. What we learn from this is that this technology has the potential, in the few domains in which it has been scientifically studied, to make us all dramatically more efficient.

And you might say, well, that's the way of the world. We've been enjoying technological change ever since the Industrial Revolution, or even before it. Wasn't the steam engine terrific, along with electrification and the invention of computers? And yes, all of those were wonderful technologies. What's different about this technology? Well, what's the same about this technology is that it's going to transform the labor market.

What's different about this technology is who it's going to affect. These previous technologies have all basically substituted for physical labor and blue-collar work. What's different this time is that this is a machine of the mind; it shapes cognitive work. Just as you saw automation fundamentally transform the labor market for blue-collar workers, this technology will impact cognitive labor as well.

Let's fast forward. We're going to see AI transform the labor market for white-collar workers this time. This makes it fundamentally different from any of our more recent technological revolutions. Okay, so we know what it does. This idea that it affects cognitive work gives us some sense of who it's going to impact.

Let's dig into that a little deeper. What I'm trying to do is think about the range of all the jobs that exist in either the Korean or the American economy. There are some jobs where AI is going to have effectively no impact. AI can't automate any of the tasks. Think about a professional athlete.

Think about a cook, a dishwasher, a floor layer, or a meat packer. None of these individuals will have their jobs significantly affected by AI. At the opposite extreme, there are jobs that involve tasks performed minute to minute, which are essentially the same tasks that AI can now accomplish faster, cheaper, and with higher quality. A prime example of this, and my apologies to whoever is translating me in this room right now, is translation. Computers are now remarkably proficient at translation and incredibly cost-effective.

But it's not just that, right? Copy editors are the same. They are. If any time you're doing this, doing written work, put it into the computer, and you'll get something better coming out. You can think about other things where it's about taking large bodies of knowledge and summarizing them, such as financial analysts, tax preparers, and so on.

So, those are the extremes. Where does the workforce in industrialized countries lie? Approximately 80% of American workers are in occupations where at least one-tenth of their tasks could be automated by AI. This indicates that AI is going to have an impact on almost all of our lives.

Fifty percent are in occupations where AI could effectively perform at least half of their job. These are the individuals who need to be concerned that the computer may be coming for their jobs. As a result of this and the nature of the tasks involved, AI will have very uneven effects across sectors. You can consider exactly where you would expect these impacts to occur.

About the legal profession, paralegals often have to read through and organize boxes of documents. Well, AI is extraordinarily good at that, and so that's a sector that is being transformed as we speak. At the other end, building cleaning and maintenance is something that AI is not particularly useful for, and so that's a sector where AI is not going to have a big effect. So that's the microeconomics. What is AI going to do? It's going to make some of us enormously more productive. And the question is, which of us?

Here is the text reformatted with correct capitalization and spelling: "Which tasks are cognitively focused tasks? What I want to do now is move to the macroeconomics of AI. What can we do with what we've just learned to see what the broader effects on the economy are? One way to think about this is to view the macroeconomy, the entire economy, as simply being a collection of tasks. It’s important to note that I am not an economist."

It's true that right now I am a speaker. I'm speaking; that's my task. This morning, my task was analyzing data and so on. If you consider it this way, one estimate suggests that AI could transform 20% of all tasks. AI may not be a very good cook, but it will be an excellent translator.

By the way, let me be clear. What I'm about to do is give you a straightforward accounting right now of the likely macro consequences. It will suggest that they are pretty small. Then, I will provide a more holistic accounting, as economists love to present arguments with "on the one hand" and "on the other," where I will suggest that the consequences could be quite a lot larger. However, I wanted to start with the very simple arithmetic of this first.

I hope it's not too early in the morning to do math, because we're about to do math. If 20% of all jobs could be transformed, one estimate is that only 23% of those jobs, or those tasks, could be automated in a way that's cost-effective. This conclusion comes from analyzing past efforts at automation. Just because something can be done doesn't mean it's cost-effective. Thus, it suggests that 4.6% of all tasks will be done by AI.

Now, if you automate a task, one estimate suggests that it would yield a 27% labor saving. This means that if we automate 4.6% of all tasks, we will save 1.2% in labor costs. However, labor is only 53% of all costs in the economy. Therefore, this indicates that if we implement AI, this estimate suggests we will achieve a 0.66% reduction in production costs.

Let me say it another way. That's the same as saying we will get a 0.66% rise in productivity. Let me be clear about what this means. This isn't saying productivity growth will be 0.6 percentage points higher every year. This is actually a forecast that indicates over a decade, it will only be 0.6 percentage points higher on average each year.

This is actually a forecast that indicates over a decade, it will only be 0.6 percentage points higher. That's barely enough. I want to be clear, this is not my forecast; this is from Daron Acemoglu at MIT. Now, this is just a productivity boost.

Now, that's not the whole story, because in order to implement AI and take over those 4.6% of all tasks, we'll need to invest in AI. Therefore, we'll require a proportionate increase in the capital stock. The term we use for that is capital deepening. It refers to the investments that businesses will need to make in AI. If we need the capital stock to be 0.66% larger and the capital output ratio to be 0.73, that adds a little bit more to the total effect.

And that suggests, all told, that over a decade, GDP, which is the measure of average output but also total income, will only be 1.1% higher. That's the Azure Mogulu estimate. Now, you might think we've figured everything out. Permit me, for a moment, to get a little more optimistic instead. The first change I'm going to make is that instead of saying only 20% of tasks are automatable, let's say 25%. That's not really a big deal.

But now I'm going to show you the estimate that comes from Goldman Sachs. The claim that only 23% of these possibilities are worth investing in arises because you think that automating tasks is expensive. In fact, ChatGPT was very expensive when it was first launched. However, the price they charge now falls roughly 50% every 6 to 12 months. The cost of the computing power needed is plummeting at an incredible rate.

So maybe this suggests that all of these tasks can be cost-effectively automated. Now, remember I said the Azure Mogulu estimate indicated that whenever you automate a task, you'll achieve a 27% labor savings. However, I presented several case studies at the beginning that showed even higher savings. If you take the average of those case studies, you could say that automating a task would yield a 36% labor saving. It is still true that labor only accounts for 53% of the total cost, but that now shifts from a small impact on productivity.

In the previous slide, it was 0.66% to 4.7% over a decade. This is now becoming quite interesting. Remember, if we want to invest enough to take advantage of those opportunities, that means we will need to invest even more in AI. Therefore, through the capital deepening channel, instead of raising the capital stock by 0.66%, we will need to raise it by 4.7%. Once we consider that channel as well, we are talking about AI increasing the level of GDP by approximately 8% over the next decade.

That's kind of interesting. It's not transformative, but, gee, it's worth thinking about as well. That's where we're at so far.

Let's go further into the macroeconomics. What I want to do is ask what's missing from my analysis so far, because this is where the optimism might really lie. I've presented a somewhat straightforward, almost accountant's way of thinking about things. However, we economists tend to be a bit more creative than that.

So the first thing to notice is that everything I've been talking about so far has been the cost savings from having AI displace workers. We’ve only been discussing the effects on the set of tasks that get displaced. But remember, that's a very small share of all tasks. Therefore, there are a lot of people—hopefully all of us, you and I—who are still going to have our jobs.

If we still have our jobs and we have access to AI, we're going to be incredible at what we do. This is a large part of the population that might become significantly more productive because we use AI in our work. For instance, I utilized it while putting this slide deck together, making the process a little more efficient. We have conducted some fairly careful studies on what happens when firms adopt AI, showing that the workers they retain and the work they continue to do become considerably more productive over the next few years.

And so it's not just about those who are replaced, but it's also about supercharging those who remain, and that's another really important point. The second thing that we've left out so far is that when you have transformative technology, the economy itself transforms. I don't have a really good way of making this point, but let's look at what the last 80 years has done to the structure of work in the United States. We reallocate workers from unproductive tasks to more productive tasks. When I look at employment by occupation from 1940 to the present, many occupations have shrunk.

For instance, farming has shrunk. This is actually because farming has become incredibly productive; as we invented machinery to do the work on the land, we need fewer farmers. We saw a decline in many labor-oriented occupations. Those jobs became less significant because of mechanization and factories. Similarly, some occupations will likely shrink as a result of AI.

Over the past 80 years, we have seen many occupations grow, including the rise of the managerial class, the rise of white-collar work, and the rise of cognitive work, if you like. Additionally, new occupations have emerged. It is absolutely true that word processors transformed the office, resulting in many fewer typists but many more software engineers. Consequently, we see new occupations emerge that complement or are associated with the new technology.

And so much of this happened as a result of the rise of computers and automation. I expect, although I’m not sure exactly which occupations will emerge, that new occupations will arise in the new economy in which the young folks in this room will eventually be working. So that's the second thing we've left out so far: this reorganization of work. Let me tell you about something that, if you really want to get excited—if you really want to get an economist excited—it's innovation. Because, at the end of the day, the thing that drives economic growth more than anything else is innovation—new ideas and new ways of doing work.

There was a time when silicon was only ever used to build sandcastles. Then we discovered a way to use silicon to make computer chips. That's a transformative new idea, and all of our economic growth is fueled by new ideas. So, what if AI can simply make our researchers, scientists, and innovators more productive?

Make our innovators 20% more productive. This would raise the rate of productivity growth, and hence economic growth, by 20%. So if economic growth was normally 2%, it would rise to 2.4%. You might say, "Well, that sounds kind of dull, from 2 to 2.4," but the thing about effects on growth rates is that they compound. As a result, 30 or 40 years later, you end up with GDP being 30% or 40% bigger. Therefore, if AI turns out to be a good innovator or a good scientist, or if it helps us become good innovators or good scientists, then there could be truly transformative effects on output.

So, it's those extra effects that then lead to—remember, we started with the sort of pessimistic arithmetic of AI, that it would only raise GDP by 1% in a decade's time. However, if you think about these other effects, here we have estimates from PricewaterhouseCoopers, Goldman Sachs, McKinsey, Accenture, and JP Morgan. You’re starting to see a similar story, which is that we have expectations that, in a decade, AI will raise output by roughly 12% on average across these estimates. Look, this is something I don't think it’s worth being particularly precise about. The question is, will AI raise output by 1%, 10%, or 100%?

And here it feels like the answer is going to be something close to 10% in a decade's time, maybe far more in the decades after that. Now, maybe this is the sort of optimistic view: that within 10 years, we’ll see quite substantial economic growth as a result of AI. Let me take a step back, though, and share with you something a little more pessimistic. What we know from past major disruptive innovations is that they often don’t have much of a positive effect on the economy that you can see for years, and sometimes decades.

That's because when you have a new technology, it's not enough to simply have the new technology; you can only realize its full potential once you've reorganized the way we do work, redesigned business processes, and restructured your organization. Our organizations are often slow to change. The invention of the steam engine in 1769 really kicked off the Industrial Revolution, but we didn't see a rise in productivity growth for 60 years, during which we experienced stagnating wages. The reason for this is that we had to reorganize how we do work in order to take advantage of this transformative technology.

Electrification didn't yield a large rise in output for another 30 years. This was because we needed to change our factory layout to take advantage of the new power that could come to us through electricity lines. For many of the last 30 years, there was a puzzle. We had the invention of the personal computer, which led the great Robert Solow, who won the Nobel Prize in Economics, to say that one can see the computer age everywhere except in the productivity statistics.

Eventually, we did see a rise in productivity, probably throughout the 1990s. This indicates that the PC ultimately changed how we work, but it took a 15-year lag. One of the big questions now is: how long will it take for us to readjust our work methods to take advantage of AI? One perspective is that perhaps AI will be really easy to integrate, as software is not physical.

It's just bits and bytes. You don't need to reorganize the factory; you just need to download it or click on the right website. The way we organize software now is through APIs, which are modular and connective, allowing you to change one part without breaking everything else.

And the amazing thing about AI is that we can talk to it using natural language. You don't need to learn Fortran or C++. You can speak to it in English, and that may speed adoption. There are still barriers that might slow things down. There's obviously a chip shortage right now.

There's the possibility of energy bottlenecks. As we discuss the political economy moment, there is a chance of regulation as people who are concerned about this new future work to slow it down. The big question is: When will AI change our lives? I think that's a much harder question than simply asking, "Will AI change our lives?" If you want to be really pessimistic, you could argue that we should look at where we are right now.

3.9% of American businesses had adopted AI by late 2023, which indicates a very slow rate of adoption. While many are discussing the potential of AI, only a small percentage has integrated it into their operations.

If you listen to an earnings call of any Fortune 500 company, they are all talking about it. However, this suggests that there are still many businesses that aren't considering the tremendous gains they could achieve by adopting these new technologies. As you think about the effect of AI on our economy, I have tried to provide you with a systematic way of understanding the consequences. However, I also want to acknowledge, and perhaps warn you, that the stories one hears can differ depending on whom one talks to. So, here I have created an axis.

You can think about AI as being just another innovation, all the way through to it having the potential to transform everything. I think as a rough generalization, when you talk to economists, we're kind of excited about it. However, we also thought the internet was cool, we thought PCs were cool, and we thought that indoor plumbing was cool.

And we thought that light bulbs are magnificent, right? They allow you to work at night. Many innovations have been pretty cool, and this is another one that's quite impressive. However, if you talk to people from Silicon Valley, the technologists, they will give you their utterly excited views about the world. I'm not here to tell you who's right.

I do know that if you're interested in the question of how AI will transform the economy, you should probably talk to some AI experts and economists. However, this will at least provide you with a lens for interpreting what you hear. Now, let's move from the macroeconomic side of AI to the political economy of AI. My goal here is not to overwhelm you with facts, but rather to offer some systematic ways of thinking about these problems, which I find tremendously helpful.

So imagine you're going to work every day, and then for Christmas, your partner gives you a robot. The robot can do everything your boss asks you to do. What a wonderful Christmas present! So every day, you just send your robot off to work and stay at home.

You do more fun things. You go for a swim. You run. You see friends. So, as you think about this robot—of course, this robot that your partner gave you—I’m really thinking about AI. Are you better off as a result of this?

Absolutely. You're seeing your friends. Is your boss better off as a result of this? Absolutely. You're seeing your friends.

Is your boss better off? No. But are they worse off? No. The work is getting done either way, right?

I want to take the exact same story, but instead of your partner giving you the robot, they give it to your boss. So now your boss owns a robot that can do your job. It's the same robot. What’s the boss going to do to you? Fire you.

So now, are you better off? Absolutely not. You just lost your job to this robot that, for some reason, your partner gave to the boss. Is your boss better off? Absolutely.

Because they're getting all the work done, but they don't have to pay you a wage. Once you notice this, it changes how I see things. The problem is no longer the robot. We don't have a robot problem. We don't have an AI problem.

We have an ownership problem. The only thing that changes between the happy scenario and the sad scenario is who owns the robot. I believe the same applies when thinking about AI. Let's complicate it a little. Now we'll move on to discussing AI.

Imagine that there are many competitors, each selling AI models that you could use at work. OpenAI does, Google does, Facebook does, and so on. Because the cost of computing is so low, they sell it for almost $0. What’s going to happen is your boss will buy that and replace you. Your boss might purchase it for just one penny because it's so inexpensive and let you go.

So, are you better off? No, you're worse off. Is your boss better off? Absolutely, you're worse off. Is your boss better off?

Absolutely, because they're able to buy your services for virtually zero dollars, as that's the price of AI in a competitive market. Now, let's complicate it one more way. What if instead of many firms competing to provide AI services, there is a monopoly? OpenAI is the only group in town that does it. So now they could sell your boss an AI.

Say your boss pays you $500 a week. What they'll do is figure, "Well, I could sell this to your boss for AI." So, if your boss pays you $500 a week, what they'll do is think, "I could sell this to your boss for $499 a week." And your boss will buy it.

So now what's going to happen is that your boss is not. Are you going to be better off? No. Your boss will buy the robot for $499, and you will get fired. Is your boss better off? Your boss is $1 better off.

Like, who cares? Is the monopolist better off? Absolutely. They are selling a product that costs them almost nothing for $500 each. This now indicates that we don't have a robot problem; we have a competition problem.

And I think this is really, really important if we consider how to divide the spoils from this forthcoming technological revolution. You don't have to think too hard to realize it's not just a problem if OpenAI becomes the monopoly provider. If OpenAI needs to buy chips from Nvidia, and Nvidia is the only company that sells them, then all of a sudden you're going to face the same sorts of problems, but it will be Nvidia that gets rich instead of OpenAI. So, competition policy is going to be really, really important. Let's move on to thinking about your life and your career.

We've covered a lot of territory. How do we make this directly useful for you? I want to apologize; many of you have quite a bit of experience with this, and I will be pitching this at a level more suited for those of us who are new to thinking through these issues. So, when people first try out AI at their jobs, maybe you're at work and your boss says to write a memo.

You go over to the ChatGPT window and tell ChatGPT to write the memo. I hear two different reactions. The first one is, "This can do my job." The second reaction is completely different: "This can do my job."

Oh, they sound similar. Sorry, let me try again. The first is, "This can do my job." The second is, "This. Can do my job."

Both are true. One is exhilarating, and the other is terrifying. What I want you to do is move beyond this sharp dichotomy and realize that the best answer is working together. This can help me do a better job, right? So the idea here is that collaboration can enhance our efforts.

Right? So the idea here is an old one in economics. Any technology, including AI, can either be a substitute for you and your work, and if it's a substitute, it will displace you, or it can be a complement, which is something that makes you smarter, more productive, and faster. Therefore, the obvious advice at this point is that you want to try to ensure that your skills are complementary to AI rather than easily substituted by it. I think that's the most important challenge to keep in mind.

As you do that, you'll need to think about what that means. I want you to understand the opportunity here. We've moved from a world rich with computer code. We have had computers for decades now, and they can take explicit instructions to transform inputs into outputs.

The problem with computers is that tasks must be written in code, which means they must be codified. This is only useful for codified languages and knowledge. We're now in a world with AI. With AI, it infers instructions from many examples. Think about handwriting recognition.

You don't tell it that this curly loop and that curly loop make this particular character. You just let it look at lots and lots of characters and let it train itself. That's what a neural network does. And so, because it can infer instructions, it can perform the task even when there are no instructions. That then means we can use this for tacit knowledge.

Tacit knowledge is the kind of experience that you and I have gained through our careers, for which we are compensated. Well, AI can now perform and utilize a lot of that. As you consider opportunities for using AI, it's important to recognize the shift from codified knowledge to tacit knowledge. When thinking about the future labor market that young people will be entering, it's essential to invest wisely in your skills. Remember, there are some jobs that will be automated, while others are impossible to automate.

You could roughly think of the skills that are definitely going to be automated, and those that are currently being automated, as programming and writing skills. Large language models write remarkably well, better than many people. So, don't invest there, or be thoughtful about how you invest in that area. On the other end, there are a set of occupations that are going to be completely unaffected, particularly those involving physical labor.

The problem with physical labor is that we don't get paid very much for it. However, the parts that AI is not yet, and I think for some time won't be particularly good at, are science and critical thinking. So, I believe that's where you should focus your investments—consider retooling for the second half of your career or deciding what to study when you go to university. I also want to suggest an attitudinal change. This part of the talk is humorous, but it's because I've spent a lot of time discussing with workers what they should do in the next stage of their careers.

And a lot of them spend their time telling me all the things AI can't do. And there are many things AI can't do. But I think that's the wrong attitude. Because if you want to be successful over the next couple of decades, your task is not to figure out what AI can't do by listing its shortcomings, but to figure out how to help AI do those things. People complain that AI hallucinates.

It does, but there are settings within AI that can effectively eliminate that hallucination. People complain that when you ask AI to write a memo, it tends to be boring and formulaic. Yes, but a good prompt can change that. If you instruct the AI to write like Hemingway, all of a sudden you'll have muscular prose, or you can ask it to write like your favorite author. You might worry, and this is true by the way, that AI makes very simple math errors.

And it does. But AI, large language models, can now communicate with statistical and other programming languages, like Python. Here’s my favorite example of this, by the way. If you ask AI, is 9.2 bigger or smaller than 9.11? Think about that for a moment.

You know the answer, right? Many large language models will actually say that 9.11 is bigger. The reason is that they've been trained on text, and they're looking at software. Version 9.11 comes after version 9.2. This is a simple example of how learning through natural language can lead to terrible math errors.

As a result, you could complain about that, or you could realize that these models have ways of succeeding despite it. That's when you can tell it to go and use Python, which doesn't have the same training to solve this problem. It will write code and run a little program for you. Some people say that AI lacks creativity, but there are actually some very good experiments on this.

And they actually show that using AI can lead to better and oftentimes more creative ideas. We like to think of ourselves as particularly creative, but we're not necessarily more so. Additionally, some people argue that computers lack empathy, claiming that AI will always be inadequate because it lacks this essential quality.

First of all, you say, "Please answer the following with empathy." That helps. And I'll tell you the second thing: late at night, I get tired and I get cranky, but AI never does.

And so late at night, AI is going to be more reliably empathetic than I am. Again, the point here is to try and figure out ways to use the technology to succeed rather than focusing on the ways that it fails. So my final piece of advice, and it's a very simple piece of advice, and I do this every day, is given the way the world is changing, what I would do is every morning open up one new tab that has ChatGPT or your favorite engine in it, and every day test drive it for one new task. Three-quarters of the time, you'll discover it's amazing, and a quarter of the time, you'll discover it's terrible, but the set of tasks that you could have done cheaply, at high quality, and incredibly quickly and productively will increasingly expand. The favorite task that I came up with is that I recently spent a couple of hours organizing a family holiday.

We went to Tasmania, which is in Australia. However, it turns out that if you simply ask a large language model to organize a family holiday for you, it will try to determine which days you should go where. That's the kind of task it's ideal for. Thank you.