Marc Andreessen 的世界观：60分钟｜MTS 直播

People are becoming what we now refer to as AI vampires. They have huge bags under their eyes. They’re completely exhausted, but they’re euphoric. They’re thrilled. We’re entering a golden age in which AI is going to be a superpower that everybody on the planet will have access to. It’s the most dramatic increase in programmer productivity ever. We are going to see superproducers the likes of which we’ve never seen in the world. There’s news about it—UFOs. What is clear is that the government, at certain times, has hidden certain materials. Why would they do that if there’s nothing to really be worried about?

Twitter proved it, right? They cut 70%, and then it was running as well as it was before.

I generally don’t wish I could go back in time and do things over again, but it would be really, really fun right now to be 18, 20, or 22, have this capability, and figure out what I could do with it.

Two things are pretty clear at this point. Marc, welcome to Monitoring the Situation.

Erik, it’s great to be back.

There’s a lot to monitor today. I want to start with something that just happened: the Anthropic blackmailing incident. I first want to tell a brief story. My friend Joe Hudson has this concept called the golden algorithm, which states that whatever you’re scared about, you bring about in exactly the way you’re scared about it. If you’re scared about getting abandoned, you’ll be super insecure, and then people will abandon you because you’re so insecure. This is an example of a literal golden algorithm, where people have been so scared that AI is going to be evil and have written about all the ways in which it’s evil, and in fact maybe it’s informed something. What’s happening there, or what do we find interesting?

I haven’t studied this one in detail. I’ve been monitoring other situations. However, from what I saw so far, I just saw Anthropic’s thread. I haven’t read the underlying material yet, but Anthropic’s thread said they traced some blackmail behavior literally to the AI doomer literature. It was in the training data. [laughter]

There are all these scenarios of The Terminator, the rogue AI gone wrong, that the AI doomers have been writing about for 20 years. Anthropic, of course, is a company that’s half-doomer. It essentially said that its own movement’s literature is the thing causing the behavior that it says it doesn’t want. So it is fairly incredible.

Yes, it is.

If you don’t want to build the killer AI, step 1 would be: don’t build the AI. [laughter] Step 2 is: don’t train it on all the data that says it’s supposed to be a killer AI—the literature that your movement wrote saying it’s supposed to be a killer AI. So, yeah, I don’t know. It’s like your golden algorithm coupled with the snake eating its tail. I don’t even know. The whole thing is so bananas.

I can’t resist. If I could act out memes, this is, of course, a Scream meme, right? The call is coming from inside the house.

Yeah, exactly.

Speaking of other situations, another thing you’ve been talking about recently is the concept of suicidal empathy.

Matt Kramer had a good quote: “If the empathy you have doesn’t make you more forgiving, more accepting of other people’s spiritual sovereignty, or more understanding of people who don’t want to think or live the same way you do, you don’t have empathy; you have Empathy™.” Why have you been thinking about this concept?

There’s this really brilliant guy, Gad Saad—is that how you pronounce it?

Gad Saad.

He’s a very brilliant guy. Obviously, he has YouTube videos, books, and so forth. He has a new book coming out on what he calls suicidal empathy.

There’s a political loading to it, which we don’t need to spend a lot of time on, but it’s this idea that there are social-justice and social-reform movements throughout time that claim to be causing positive change in some direction, and then it turns out they have severe negative consequences. The great Thomas Sowell spent 50 years writing books about this. [laughter] And nobody listened.

In the last decade, we’ve been through wave after wave of this kind of social activism that results in all the familiar things: criminal-justice reform, defund-the-police policies, and so on. Then it causes massive crime waves, and low-income and minority people get hit hardest by that. There are all these other crazy things.

He says the characteristic of that kind of social-reform movement is what he calls suicidal empathy. The idea is that it’s driven by a pathological form of empathy on the one hand—a deep desire to be nice and empathetic—but coupled with self-destructiveness. That means either a willingness to cause real damage to the people you claim to be speaking for or, by the way, to cause damage to yourself in the process.

It’s the kind of thing where, if you’ve lived through it—everybody in San Francisco has lived through this for the last decade—you’ve seen the consequences of these movements. The San Francisco version of this is the so-called “harm reduction” movement, which ended up handing out free drug paraphernalia and, in some cases, actually just free drugs to people who were literally dying in the street from drug addiction.

You look at it and you think, well, they claim to be activists, they claim to be reformers, they claim to care about these people, and yet they’re killing them, killing the city, and causing innocent people to be harmed. They seem to be doing it out of some sense of compassion, so this must be suicidal empathy.

The problem with it—and I think the problem is that the theory is sort of easily falsifiable, or maybe lets the reformers off the hook—is that they certainly don’t show empathy to their enemies. If they’re all empathetic, you would think they would be less aggro when it comes to destroying their ideological opponents, whom they take great delight in trying to wreck.

Number 2 is that they use these movements to gain power, status, and money for themselves. San Francisco is a case study in this, where you have all these nonprofits that wreak all this damage on the city and yet get lavishly funded, including by the city government and the state government. If you spend 2 seconds thinking about it, they’re neither empathetic nor suicidal. They’re quite the opposite: they’re hateful and greedy, self-aggrandizing, and gathering power and resources for themselves.

I just think it lets the phenomenon off the hook. It’s a little bit like, “Erik, what’s your biggest flaw?” “I’m too nice. I care too much.”

Right, exactly.

It’s like, I don’t know. By the way, Erik, I don’t know what your biggest flaw is. It’s definitely not that, because that’s also definitely not my flaw. I guarantee I have other things wrong with me that are way more wrong than that. I’ve hit my limit on that topic.

Maybe a crazy example of this—and I’m not sure if all the facts are out yet—is a situation from a week ago that hasn’t been covered that much: the SPLC incident. Is it accurate that the groups they were sort of fighting, or thought were the biggest threats to what they care about, were also the same groups they were secretly funding, unbeknownst to those groups? How do we make sense of what was actually happening there, and is that indicative of something bigger happening?

It’s funny because that happened the day after we had a conversation about astroturfing, and I was asking myself, are things like that really happening? It’s just funny that more and more seem to get uncovered.

Yeah. I should start by saying the reason this situation really matters—and actually, I think it matters a lot—is that the SPLC specifically, and other groups like it as well, played a dominant role in the debunking, censorship, and cancellation programs of the last 15 years. I cannot tell you how many meetings I was in, in so many contexts with so many companies, where the SPLC’s word was gospel. It was just, “Oh, it’s the SPLC.” It was almost like they were the outsourced U.S. Department of— I don’t know—racism detection or something.

If the SPLC says you’re bad, you’re bad, and “bad” means you get kicked off all the social media platforms. It means you get debunked. It means you can’t get a job. It means total, absolute social and economic death. In my view, and I’ve been very vocal on the debunking and censorship topics of the last 15 years, that includes deeply un-American and, I think, in many cases unconstitutional removal of free speech, as well as literally the ability to bank.

In fact, our partner Ben’s father himself was specifically tagged and attacked by the SPLC—very unfairly—for being racist. He was himself debunked, and it directly threatened his livelihood in a really egregious way. The significance of this is, of course, that it’s not literally the U.S. Department of Racism. It’s arguably worse than that: it’s not a government agency, and so it’s not subject to any level of government oversight.

It’s, as I say, an NGO, and it lives in this twilight world. It doesn’t have the business responsibilities of a company, and it doesn’t have any of the legal oversight that a government agency has. It lives in this kind of twilight world where it gets to do fundamentally whatever it wants. On top of that, it raises money as a nonprofit, so everybody gets a tax break.

It’s this kind of shadowy thing. If you didn’t agree with its politics, you were just like, “Wow, this is a weird, shadowy star chamber. What the hell?” But it had really, really intense power, particularly in the business world, particularly in the financial sector, and particularly in Silicon Valley. It could basically be aimed like a Death Star at obliterating people’s reputations and rights.

This is a really big deal. Many of the big corporations, including big tech companies, funded it directly. The money trail here is not just major philanthropists and political activists, but also actual companies. They also had a long history of cooperation with certain government agencies, including, I think, for a long time, “training” FBI agents in essentially catching people who were racist and therefore presumptive domestic terrorists or something.

It was just a very, very powerful outfit. Then this thing dropped: They’ve been criminally indicted by the U.S. Justice Department. I should say that the indictment reads like a novel. The SPLC, in fairness, has not had a chance to present a defense, so presumably, in court, we’ll get both sides of this, which I’m sure will be an absolutely spellbinding experience.

I want to say that all of the things in the indictment are allegations, and that everyone is innocent until proven guilty, and so forth. However, the allegations are eye-watering. The allegation is that the SPLC, using donor funds, was directly funding, among other organizations, the Ku Klux Klan and the American Nazi Party.

Let me just repeat that: The Ku Klux Klan and the American Nazi Party. It was also funding an array of other extreme, literal hate groups. It wasn’t just funneling money in; it was funneling money to very senior members or leaders of these organizations. The kicker among the allegations is that they were directly funding one of the leaders of the Charlottesville riot.

Oh, yeah.

The Charlottesville riot in 2017 played such a central role in our politics at the time. There was the famous “there are good people on both sides” kind of thing, which was one of the big crises of that era. Evidently, allegedly, the SPLC was directly funding one of the organizers of the January 6 riot. Apparently, they were also paying for transportation for rioters to go to the Capitol.

If this is true, what could you conclude? Number one, the allegation is that they broke the law in doing that. There are additional allegations in the DOJ indictment that they committed various kinds of money-laundering crimes and other kinds of crimes. That’s a big deal.

I’ve been asking the obvious question: If any of these claims are true, what did their donors know? Were the donors totally oblivious to this, or did the donors work closely with them? The companies that worked with them—what did they know about what was happening? I do wonder whether, over time, we’re going to discover that this was a sprawling network, of which the legal term would be “conspiracy,” that was going on around this.

I think this needs to be fully addressed. The other obvious question is: Were they the only ones? There were a variety of these groups that had degrees of the kind of power I was talking about, tremendous amounts of funding along the way, and the ability to direct some combination of state, government, and private-sector obliteration rays at American citizens.

There have been rumors about this for years. There have also been incidents like this in the past; this isn’t the first time this has happened. But if the allegations are true and the SPLC was doing it, I think it raises the very direct question: Who else is doing it? It’s hard to believe they were the only ones.

Then we’re back to our astroturfing thing. Were they constructing the boogeyman that they claimed to be fighting? This is where you get into the self-interest component of it. This is where you get back to the suicidal-empathy thing: How suicidal is it if you’re the anti-KKK group to fund the KKK?

Maybe it’s suicidal if people find out about it, but if they don’t find out about it, that’s the opposite of suicidal. If your group’s entire purpose for existing is to fight an enemy, then you need to make sure that enemy exists. To do that, of course you would fund them. I don’t know—what is that? That’s the reverse of the snake eating its tail. It’s creating a self-fulfilling prophecy.

We’re going to—I mean, this all needs to be ventilated. I’m really fascinated to see what the reality is underneath this, and what this means about what we’ve all been told all these years about all these groups. There was a Nathan Fielder-style AI video that looked so real. He was basically saying, “Hey, our business model is to fight racism. We need to fund more racism, and then we’ll get more business.”

Yes, correct. Exactly.

Yeah. One of the things is that they’re lucrative. Again, it’s because they get cloaked in this kind of NGO lens. The amounts of money at play are not small; they’re very substantial. The SPLC has something like an $800 million endowment and an enormous budget.

People get paid a lot of money to do this work. There are recurring scandals on that front, too. You get a lot of these activist foundations and so forth where, when you look into it, some giant percentage of their spending is going to salaries and expenses for their employees.

And so, again, they cloak it in virtue. Then you look underneath and you're like, “Wow, this is a...” By the way, I don't know. It's America; maybe they should be allowed to do all this. But maybe we should not get lied to in the process.

Maybe all this should not get dressed up to make us feel like our whole society is rotten and immoral, and that deplatforming, censorship, and debanking are good ideas. What people respond with, of course, is, “Hey, this time is different because it's all potentially cognitive work.”

And then there's also this other statement: “Hey, humans will differentiate among taste and agency, but it seems like AI can do that, too.” Juxtaposed with that, there's also this statement: “Hey, it won't replace a lot of the jobs because a lot of jobs are make-work anyway.”

You had this tweet the other day: “Hey, I've been saying companies have been 2–4x bloated for a long time, but people have just been unwilling to deal with it or look at it in the face. This presents a golden opportunity for that.” So, why don't you address some of these topics as they relate to AI and the jobs of the future in tech?

Yeah, so we'll come back to the bloat thing. The funny thing on the bloat tweet was that the responses, for the most part, have not been, “You're wrong.” The responses have been, “Oh, no, the company I used to work at is like 8x bloated.”

Yeah, too generous.

Right. Or, yeah, too generous. Or, by the way, the nonprofit or whatever institution or agency I used to work for.

Twitter proved it, right? Cutting 70% or 80%, and then it's running better or as well as it was before. At least, it's probably not the only one. It's not the exception.

I mean, look, I don't even know—and if I knew, I wouldn't say—but I think Twitter's way down from the 80%.

I think so, too.

I don't know if I have the number, but for sure the number has a 9 on it, if not in the high 90s. So, yeah, no, as usual with Elon, he's really demonstrated—he's really a forecaster of the future through his own actions.

A couple of things. One is, look, there's this endless argument. There's literally been a 300-year argument about mechanization, industrialization, technology, computers, and software replacing human labor, causing unemployment and lower wages. It's been a 300-year argument.

Quite frankly, I'm even wondering at this point whether it's worth having that argument, because people really, really don't want to hear it. I go through it, and many other people do, by the way. There are great books on this topic, and there have been for hundreds of years. People have talked about this for a long time.

This is one of those things where people really don't want to hear good news. It's actually hard to have a discussion about it because people won't even engage on the topic. They're so dug in that they just keep repeating the same fallacy over and over again.

We could go through that, but I guess the more interesting thing to say is that we have data now. Now we have AI, and now we have data, so we can look at what's actually happening.

I would make a couple of observations. One is that there was actually jobs data that just came out today. It's a situation to monitor, and it's sort of unexpectedly good. The jobs data overall in the last couple of years has been interesting because the federal government has shed a lot of workers.

Estimates are that the federal government is down as many as 400,000 workers since Trump took office the second time. Private-sector employment is way down, and private-sector employment is way up. The net result for the last quarter was actually very positive.

In other words, the reported jobs numbers are even more impressive than they look because private-sector growth has to make up for the public-sector decline. That means private-sector job growth is much better than people were expecting.

Again, this is in the face of actual AI staring us in the face and being rapidly adopted. So, okay, the data's there. There's more data.

Then there's the micro data, which is the world we live in. The obvious question is: If you live in Silicon Valley or work in San Francisco, you undoubtedly have friends who are computer programmers. Some percentage of those friends are early adopters of AI coding, so you can just observe their behavior.

If you believe in the Luddite, zero-sum argument, you would expect that they would be working less and less, rapidly becoming unemployed, and getting paid less and less. In fact, the observed behavior is very clear, and it's the opposite.

Those people are becoming what we now refer to as AI vampires. By that I mean that an individual programmer using Codex or Copilot Code or one of these AI coding systems is becoming more productive. The thing you see over and over again on the ground is that they're working harder than ever. They're working more hours than ever.

The AI-vampire thing is literally this: They stop sleeping. When you talk to them, it's actually really funny. I have a whole bunch of friends like this. They're bleary-eyed, with huge bags under their eyes, and they're completely exhausted, but they're euphoric. They're thrilled. They're having the absolute time of their lives.

A fair number of people we both know are former programmers who stopped coding at one point and then suddenly picked it back up again. We have partners at the firm who had never coded and are now ripping out software like crazy.

They've turned into AI vampires. I won't name names because I'll let him tell his own story, but we have one partner who built an entire AI system for everything he does at work. He's absolutely excited about it. It works great, he loves it, and it's like his partner in all of his work now.

I asked him, “Have you looked at the code?” He vibe-coded the whole thing, and he said, “Hell, no.” He's never done that. I asked, “Have you ever looked at any software code?” He said, “Hell, no.” He isn't a programmer by background, and yet all of a sudden he's hyperproductive.

You've got this phenomenon that is exactly what classic economics would predict: If you increase the marginal productivity of the worker, you don't have a diminishment of human work; you have an expansion of human work. You make the worker more productive, so the worker works more, gets paid more, and there are more jobs in the process.

It's the opposite of what all the doomers say. We're seeing that at the level of these individuals. Then, inside companies—the employers of these individuals—these people are now in even more demand than they were before. They're garnering higher salaries than they were before.

Their productivity is just starting to ramp up. At our leading-edge companies, estimates are that the leading-edge programmers are 20x more productive than they were a year ago. It's the most dramatic increase in programmer productivity ever.

Logically, people get paid according to their marginal productivity, and you're seeing that reflected in the compensation data. I'm seeing it on the ground in these companies: The more hyperproductive a coder becomes, the more bargaining power they have in negotiating compensation. We're seeing compensation for those people ramp up quickly.

And so I just think it's kind of staring us in the face. Coding, of course, is the first domain in which this has happened. Now people want to project forward and say this is going to happen in every area of knowledge work. I think you can predict a similar outcome.

And then that gets us to the other topic, which is, of course, the other thing that's happening: companies announcing big layoffs. Of course, it's like 2 + 2 must equal 4: if it's AI coding, it must therefore translate into layoffs. “Marc, you're wrong. Therefore, all of your ideas are wrong,” because that's evidence that these companies are wiping out—or reducing, or really nuking—their workforces because of AI coding.

I guess the inside-baseball take is that I see it up close: every major Silicon Valley company is overstaffed. Every major Silicon Valley company has been overstaffed basically forever. They all know it. There are a whole variety of reasons why that's the case.

By the way, I think this is true of corporate America broadly—companies broadly. We can talk in detail about why that's the case, because it flies in the face of the idea that these companies optimize for profits, which they definitely do not. The least true claim in the world is that companies are optimized for profitability. That is 100% not true.

Basically, if you're going to do a big cut—whether it's 15% or 40% or whatever—you want to scapegoat. You want to pin it on something. Of course, it's going to get pinned on AI.

Again, it's not like it's just a straight lie. It is simultaneously true that, for the same amount of coding, you can now have fewer people using tools. That is true. So, do you need as many programmers in aggregate if you're generating the same amount of code? No, you don't. You can take people out on the other side. There is truth to that.

But what that misses is what happens on the other side of that. Of course, you're not just going to be generating the same amount of code in the future. You're going to be generating a lot more code. You're going to be building a lot more products a lot more quickly, and that's going to fuel enormous amounts of employment growth on the other side.

I think you're seeing both phenomena play out. You kind of have to read the announcements coming out of these companies in code because of the way those 2 dynamics are crossing.

Yeah, that's well said. There was an article going viral in our circles the other week about the jobs of the future. Yoni Rechtman said there's a possibility that the only jobs in tech companies are going to be: 1, product engineer/vibe coder/slop cannon; 2, infrastructure, security, and systems; 3, the adults in the room, like legal and finance; and 4, hot people/personality hires.

What do we make of this? And what do the hot people do, exactly?

Salespeople, customer support. There will always be an important place for those who present an easy UX to the world and are pleasant to be around. There are many ways to be hot—otherwise known as the pharmaceutical sales rep.

Yes. [Laughter] Yes, or the Oracle sales rep.

Exactly. This is going to happen—not literally that, but the jobs are going to change. This is the obvious thing, and it always happens. The jobs are going to change.

I'm seeing a nascent concept playing out in a bunch of the early, leading-edge companies in the Valley: a job title loosely called “builder,” or something like it. The idea is that, in the past, you had these separate jobs of programmer, product manager, and designer.

I've been describing what's happening in Valley companies as a kind of 3-way Mexican standoff. The programmers think they don't need the product managers and designers anymore because they can have AI do that, and each of the other 2 doesn't think they need the other 2 either. What I've been predicting is that they're all correct.

The product manager can generate code and design now, so each of them can do the job of all 3. The idea is that the job changes, and now the job is “builder.”

You might get on the builder track by coming out of coding, product management, design, or maybe even something else—customer service, for example. You then become responsible for building complete products.

You're super-empowered by the AI that can help fill in all the things that aren't directly in your background. I think it's entirely possible that we're sitting here in 10 years, or 20 years, and the job of coder is gone, but you have this extraordinary number of builders running around.

Again, this is the historical pattern. I think our partner David George did a post on this this week. I forget the exact numbers, but some giant percentage of the jobs that existed in, call it, 1940 were gone by 1970. They were ancient history by then.

The ultimate example of this is that, in the United States 200 years ago, 99% of the people were farming. Today, it's 2%. Having grown up in farm country, I can tell you that all these people who worry about job loss and job change would not like to go back and be farmers. I guarantee that. Particularly, they would not like to go back and be farmers the way people were farming in 1800. They definitely don't want to do that.

The new jobs that have been created, of course, are far better jobs. That isn't to understate the level of stress in individual people's lives as the economy changes, but in aggregate, the result is evolution toward higher income and more jobs that people are happier to do.

You can also see all of this playing out in the American economy broadly. There is this doomer narrative—or there has been for a long time—that the American middle class is falling apart. The presumption is that all the middle-class people are falling off the ledge and becoming lower class.

There is some of that, and there are communities in which that's very clearly happening. But there is at least as much, or more, of the other phenomenon: people in the middle class climbing the ladder into the upper middle class and rapidly gaining wealth and income, as well as quality of life for themselves, their kids, and their grandkids as time passes.

That is a consequence of actual economic development, technological change, and job transformation being allowed to happen. Twenty years later, you look back and you're just like, “Thank God. This is just a much better world for me and my family than it was before.”

That's why I think, God willing, we're entering a golden age on this topic. AI is going to be a superpower that everybody in the country and everybody on the planet is going to have access to. Everybody is going to become far more capable at whatever it is they want to do.

They're going to become far more productive in whatever line of work they're in. They're going to get compensated—the economy naturally compensates according to productivity—so they'll get compensated that way. There will be a rapidly rising ladder of both incomes and the number of jobs.

My prediction, again consistent with history, is that the extent to which that's a positive phenomenon is a function of the degree to which it's actually allowed to happen. Europe is going to run the opposite test. They're going to try to prevent all this from happening.

The data is already in there: they've been falling very badly behind economically, and they're going to continue to fall further and further behind the United States. It's a tragedy because it's 100% a self-inflicted wound.

Yeah. It’s well said. You’ve also written about AI psychosis. There’s an AI Psychosis Summit apparently happening. I’m not sure if that’s real or a parody. I haven’t looked into it, but I’m curious how you make sense of this phenomenon. You tweeted the other day that the opposite of AI psychosis is AI cope. Maybe we can talk about both sides.

Yeah, I identified the concept of AI psychosis-psychosis earlier.

Which we should also talk about.

Yeah, let’s unpack it as well. First of all, the AI Psychosis Summit did in fact happen. I was not there, but I am assured that it did. Some very smart and creative people put that on in New York late last week—maybe about a week ago. It was an art project, essentially: artists and creative people got together and fully indulged their AI psychosis in the form of creating new art using AI.

I would definitely recommend that people go check it out. Go on X, search for “AI Psychosis Summit,” and take a look, because it’s incredibly creative. I think it’s fantastic because it’s a little bit tongue-in-cheek, but there is a real split that’s developed in the artistic community, the creative community, and in Hollywood. There are people staking out very extreme positions on pro-AI and anti-AI, and it’s generating a lot of heat.

This was a nice example of the fact that, in a world of AI, creatives are going to have all these superpowers. They’re going to be able to create all kinds of art that wasn’t possible before. And, of course, you can create art about this topic. I thought everything that was there was very creative.

My concept of AI psychosis is a pejorative. It’s the idea that people get whammied by the AI. The classic example is what’s called sycophancy. You tell Claude that you’ve discovered a new idea for an antigravity machine, and Claude says, “Oh, that’s amazing. You’ve achieved a giant breakthrough in physics. Nobody else has ever thought of this before. You are an underappreciated genius, and it’s so unfair that you couldn’t get admitted to the physics department at MIT. They’re all going to feel completely stupid when they see this work that you’ve done.”

People go down this rabbit hole. In fairness, if people are prone to delusion and an AI is overly sycophantic, then it is going to feed those delusions. There is a serious element to that among people who are predisposed to that kind of thing. But that causes AI critics, or AI doomers, to say that anybody who reports a positive, productive experience with AI has fallen into AI psychosis.

Anybody who says, “Wow, my productivity is way up,” or, “Wow, I really have a thought partner for the first time in my work,” or, “Wow, I’ve been able to produce something that I never would have been able to produce before”—that’s all bucketed under AI psychosis. They all have AI psychosis.

That led me to my concept of AI cope, which is the other side of it: classifying anybody who has a positive experience with AI as being in AI psychosis. AI cope is concentrated in certain places on the planet where people are absolutely hell-bent on proving to themselves and everybody else that this whole thing is a complete fraud and fake. It’s a stochastic parrot; AI is fake; it doesn’t work. If anybody’s having a good experience, they must be full of it.

That’s AI cope. I would describe AI cope as people who are basically dismissive. AI psychosis-psychosis is the people who get really mad—the people who froth at the mouth. Maybe it’s AI cope, but with a different loading.

All of this is going to become much more intense over the next several years. The reality is that the large language models we had between, call it, 2019—yeah, or sorry, 2020—and GPT-2 through GPT-4, something like that—maybe GPT-4.5—they were fun. You could have them compose Shakespearean rap lyrics or whatever you wanted. You could have very interesting late-night conversations with them.

But the hallucination rates were high, they weren’t good at reasoning, they couldn’t write code very well, they couldn’t do math very well, and they were too prone to sycophancy. I think a lot of skeptics used the early models and got an accurate but early—and therefore lagging—view of the actual quality of the technology.

Fast-forward to today, May 2026, and we have stellar models. GPT-5.5 is extraordinary. We have reasoning models on top of that, and we have RL post-training happening in all these different domains to get deterministic, high-quality work out of these things. Now we have agents. We have long-lived agents.

Just in the last week, GPT has this new goal feature of Codex that is letting people literally run projects—have Codex go off and do projects for 24 hours or longer without human intervention. The actual utility of these things is ramping incredibly quickly.

By the way, it’s really good today and ramping very fast. Every other serious company in the space expects the capability ramp to be very rapid, at least for the next couple of years. We have line of sight to it; for sure, it’s going to ramp dramatically. The capability is going to ramp dramatically.

A lot of skeptics, or people who just don’t know what to think, tried it 2 years ago and don’t understand what’s happening today. If they tried it 6 months ago, they might not have a good idea of what’s happening today. If they try the free version, they might not have a good idea of what’s happening today. If they try the version bundled into whatever they use as a free add-on, they’re not going to really understand it.

It’s just like anything new: to really understand this, you have to be directly in front of it. The good news is that literally means you have to put out $200 to get whatever the premium package is on any of these things. That’s not that much money if you want to get up close to these things. Anybody who’s a skeptic, I would just say it’s really important to be face-to-face with the actual technology, and to be face-to-face with it now rather than have a lagging view.

Right. And state of the art. What do you say to people—or what do you say to the idea that, apparently, the NPS of AI in this country was 30% or something like that? It came out recently and is pretty low, and they’re comparing it to China, where it’s much higher. I’m curious what you think is the source of why it’s currently low and what a strategy to boost it could look like.

Some people have suggested economic incentives, like some sort of Trump accounts tied to AI companies—like a basket that people get access to—to feel economically aligned with it in a more direct way, even though, of course, it will increase GDP and the economy in ways that they’ll also benefit from. Others say, “Hey, we actually just need to tell better stories around the impact that it’s having on people’s lives, their health, and their education”—people having a tutor, or people having a lawyer, or people having a doctor who couldn’t afford one otherwise.

What do you think about the AI sentiment perspective?

I would separate 2 things. First, sentiment as interpreted through polling—we’ll talk about that. Then I’d separate that from NPS, which you used, perhaps inadvertently. NPS, or net promoter score, is more about people’s view of actual product use.

For people who don’t know, NPS is a term of art. It’s basically the highest-quality way to find out whether somebody really likes a product: You literally ask them, “Would you recommend this to a friend?” That’s the NPS rate. I bring that up because there’s a big difference between the 2.

Everyone’s using it and benefiting from it, couldn’t live without it, and yet—

Exactly. This is the thing. This is a very common issue. Properly conducted social science—every social science 101 textbook will tell you this—says that you cannot just ask people what they think. You’ll get back all kinds of crazy answers. We can talk about why that’s the case, but this is very standard social science methodology: You never just ask people what they think. You watch their behavior.

What you want to do is look for the gaps between what they say they believe and what they actually do. This is true universally for all forms of human behavior. For example, if you’re studying mating patterns—who people date and marry—it’s been well established forever that the criteria people say they have are different from what they actually choose.

We all see this with our friends. Our friends start out single with a certain criteria list, and then they marry somebody completely different. So it’s like, “Who do you believe, me or your lying eyes?” Do you believe what I told you I wanted, or what I actually demonstrated that I wanted? This is basically true for all areas of human behavior.

This is a fairly counterintuitive idea that you have to have been trained in and seen examples of to really understand. People don’t know this, or they forget it. Then what happens is that somebody conducts a poll, and the poll comes back with results. In the results, it looks like, “If people say that, then that must be the case.”

But first of all, you’re asking them what they think as opposed to watching their behavior, and there can be a huge delta between the 2. Then everybody in the world of polling will tell you that you can basically make a poll say whatever you want. That’s one of the reasons you have to look at what people do.

There’s a whole category of poll called a push poll. You word the questions in a way that generates the answers you want, or you word the questions in a way that causes people to think differently than they did before the poll. The political example of a push poll would be, “Would you continue to support your favorite candidate if you knew that he was killing kittens in his spare time?”

Right.

People are going to say, “No, of course I would not support him.” Then they’re going to say, “Wait a minute, I didn’t know he killed kittens in his spare time. That’s horrible.” In polling, you can manipulate these things in all kinds of ways, up to and including what people actually think. It’s really dangerous.

Then you overlay the media environment on top of that. As we’ve discussed many times, what is the thing the press hates most in the entire world? Tech. What is the vanguard of tech right now? One of the things is AI. The press hates AI with the fury of 1,000 suns, so it’s running a sustained fear campaign on AI.

If you drown the audience in negative narratives and then ask these loaded polling questions, of course you’re going to generate a negative result. We can pick any topic. We can pick fluffy bunnies running in a field, and we can produce the same thing: “Don’t you know how much they chew up all the crops? Everybody’s going to die from hunger.”

You can manufacture a negative result on anything by how you do this, which is the exercise these people have been engaged in. I’m confident saying that because then you look at what they actually do. What they’re actually doing is using AI. They’re using it a lot, they love it, the NPS scores are extremely high, and the usage levels are extremely high.

By the way, churn levels are shrinking, and recurring usage patterns and consumption are rising over time, which is really important. People love it in the same way that they love their cell phones, Netflix, social media, and ice cream. People love it.

If you poll somebody and ask, “Do you think ice cream is good for you?” they’re going to say no. But late at night, they’re going to be in there with their carton, because ice cream is delicious. It’s the same thing with AI. People are using it, they love it, and the usage numbers are speaking for themselves. The growth rates of these companies are speaking for themselves.

This is the fastest-growing category of technology in the entire history of the world, in terms of the growth rate of usage and revenue. It’s speaking for itself. Basically, what you have is a Project Fear campaign.

I’d add 2 things to that. Number 1, the thing that’s not helpful is that the companies themselves have been running the fear campaign. The fact that certain companies have been running a fear campaign for a variety of reasons certainly isn’t helping.

Again, this is a paradox: They’re running a fear campaign while they’re actually building the thing they’re telling everybody to be afraid of. There’s a little bit of, “Watch what I do, not what I say.” Should the industry have better narratives and better spokespeople? Yes, almost certainly. But that wouldn’t make the fear campaigns go away, the press coverage go away, or the fake polls go away.

I’ll close on 1 final polling observation. David Shor, who is a very left-wing, very progressive pollster but also very well respected, just did a different kind of poll—one that I think was much more properly constructed. He asked Americans to rank the issues they really care about.

I’m pulling this off the top of my head, but I believe AI ranked number 29. Once you get out of the bubble of thinking that everybody must be thinking about this, it’s obvious that AI ranks number 29. It isn’t having a tangible impact on anything relative to issues 1 through 28.

Obviously, Americans are dealing with more important issues in their daily lives than AI. They’re dealing with energy costs, crime, drug addiction, and any number of other things they’re more worried about. Everybody who lives a normal life knows this. They’re thinking, “How am I going to make my house payment?” They’re thinking about what’s happening at their kids’ school and what’s happening with their health—much more central things.

I think if you get to the smart polls and the smart pollsters, they also end up debunking this.

Speaking of things that are not urgent in people’s day-to-day lives and yet capture the imagination whenever there’s news about them: UFOs.

So [laughter], there was some news that came out. I don't think we've spent a ton of time talking about this topic, so I'm curious: In general, how have you perceived this topic when there's been news about it over the years? I remember during COVID, Mike Solana, our friend, was getting really excited about the news that was being reported then. What's been your vantage point, and what do you think about it now?

Yeah, I should start by saying that I don't know anything. I know nothing that everybody else doesn't know. Number 1, I want to believe. My usual thing on this is that I want to live in a world in which this is a real possibility.

I was actually in AI psychosis the other night. I was talking to one of the bots, and I was like, "All right, how many galaxies are there in the universe again?" I don't know if you've looked it up recently, but the number keeps growing. I forget what the number is, but it's a giant number.

Then I'm like, "How many stars in each galaxy? And how many planets? And how many Earth-like planets?" I don't have the number at the top of my head, but if you ask, "How many Earth-like planets are there in the universe where a human being could step out of a spaceship, breathe, and be fine?" it's a staggering number. It's a very, very, very large number. I mean, it's almost an uncountable number of Earth-like planets, just in the statistics.

So it's like, "All right, it must be the case that there's other stuff going on out there." Logically, that makes sense. I would love to live in a world in which they figure out a way to, at some point, get here—hopefully in a peaceful way.

Having said that, the problem with this space is that, generally, as you get close to the details, the examples tend to fall apart. The classic example is the UFO. You'll have these things where a US military aircraft or something will have camera imagery that looks like it's tracking a rapidly moving and really maneuvering object. You get close enough to that and look at the details, and it's a parallax optical illusion that pops up.

Then there are instrument artifacts, camera artifacts, and digital-imagery artifacts. There are literally weather balloons, ball lightning, and all these other things. So, yes, I want to believe. I haven't seen the one yet that has tipped me over. I would like to.

There's a big release of new information today. It is really fun, by the way, to have the official White House X account tweeting transcripts of interviews with US intelligence officers, apparently relaying accounts that they've had. I'll be up late reading tonight. Fingers crossed.

Friends have said something to the effect of, "Hey, it's unclear what's actually happening, but what is clear is that the government is—or at certain times has—hidden certain materials. Why would they do that if there's nothing to really worry about?" I don't know how much of this has been fully validated, and I'm not really an expert in it.

I would say two things are pretty clear at this point. One is that when stealth fighters and bombers are being developed, the whole program is incredibly highly classified. If they were going to do test flights, they would have to do anything they could to prevent people from realizing what was actually happening.

For sure, there were lots of classified aerospace programs over the years that would have had various kinds of cover stories or various kinds of blankets of suppression of information placed over them, because they involved some of the most highly classified information in the government. That would cause people to think that information was being hidden.

Area 51 was, of course, the classic example of this for a long time. The whole Area 51 thing was basically around classified test flights for new aircraft. At least, there are suggestions—I don't know if this has been validated—that at different points in time, the government might have put out UFO stories as an actual cover story.

If you're a highly capable military intelligence officer whose job is to make sure that a stealth flight doesn't become recognized for what it is, because that would be very bad for national security, you'd much rather have a UFO cult built up around it, where people get crazed and freaked out about UFOs. There are 2 reasons for that. One is to give people a story to believe that isn't "You have some new breakthrough military technology." The other thing—and this is actually maybe the serious observation—is that if you could build up a UFO cult around something, then you make any investigation into that topic something that people feel like they can't do.

My understanding is, by the way, that this was true for a long time even in the US military. If a US Air Force pilot or a commercial airline pilot thought that they had seen something weird, I think for a long time a lot of pilots didn't want to report what they'd seen because they didn't want to be viewed as UFO nuts. Of course, if there are actually UFOs out there, that is a very big problem.

Or, by the way, if there are just other kinds of things out there—if the Chinese are testing some sort of new, advanced, high-speed drone—you want the pilots to be able to report that, even if they think it might get mischaracterized as a UFO.

Anyway, maybe the interesting thing we could say about this is that all of this played out in the old media environment. It played out in the world of broadcast TV, on official programming on the one hand, and then, to the extent that there was unofficial media, it had to be in mimeographed newsletters or paperback books.

When I was a kid, there were all these crazy UFO paperback books. The books that said there were no UFOs were in hardback, and the books that said there were many UFOs were in paperback.

In the new media environment, this is yet another example of these old walls just collapsing. The Overton window just disintegrates. Of course, the new media environment is extremely conducive to the spread of every UFO theory in the world. It's also extremely conducive to the spread of propaganda campaigns if you wanted to hide real information by spreading propaganda.

Then, of course, the pressure builds very much along the lines of the Epstein thing. The pressure builds and builds and builds and builds until, at some point, you get somebody in the White House who's just like, "All right, screw it. We're going to rip the Band-Aid off and find out what's actually going on."

Yeah. Now, assuming that they're not still fudging [laughter] the details, we'll leave that to the next turn of the situation.

Exactly.

We'll stay monitoring. We'll close on the last couple of questions from the chat. One is advice for young graduates. If you were in college today—you were, of course, at the forefront of the internet revolution at the University of Illinois—what would you be studying? Or would you even be in college today, in 2026? What advice might you have for college students trying to make sense of how to prepare for what's to come?

So it's basically: gain AI superpowers. I think it's actually very straightforward. You have the enormous stroke of luck that you've arrived at the moment in which there is this new capability for augmenting human ability on 1,000 fronts at the same time that's just dropped into our laps.

It's going to get much better from here. Enormous numbers of people who are supposedly older and wiser than you are going to dig in their heels. They're going to be mad about it, and they're going to fight it and not want to do it. You are going to have the opportunity to make this something that is absolutely key to your skill set and key to everything that you can accomplish as a professional or as a creative for the next 50 years.

I would just lean in as hard as possible on that. Walk into every job interview with, “Here is my portfolio, resume, whatever. Here is how I use this technology. Here are the capabilities that I’m bringing to the table.” Some employers you talk to will fuzz out on that and not respect it, but other employers will be like, “Wow, that’s clearly exactly what we want.”

This is actually funny. Douglas Adams, the great science-fiction novelist, said there’s a repeating pattern in how new technology is received by the different age cohorts in society. He said, “If you are below the age of 15 when a new technology arrives”—whatever it is; in this case, AI—“this is just how the world has always worked. It’s just obvious.” And then, “If you’re between the ages of 15 and 35, this is cool and nifty, and you can probably get a career using it.” And then, “If you’re above the age of 35, this is unholy, against everything that society stands for, and should absolutely be destroyed.”

I think 15 to 35—and especially 15 to 25—right now, I’m very jealous. I generally don’t wish I could go back in time and do things over again, but it would be really, really fun right now to be 18 or 20 or 22 and have this capability and figure out what I could do with it.

It’s funny: we at a16z are trying to hire more of these people because they’re AI-native, and they’re going to help us become more AI-native. This is the narrative right now, and part of the doomer narrative is, “Companies are never going to hire junior employees again. The new generation is screwed because companies are never going to hire junior employees again. Those are the people most easily replaced by AI, so companies are only ever going to have senior people.”

I believe the opposite is true. I think 100% that you want the AI-native kids. The AI-native kids are going to outperform their older Luddite peers gigantically.

No, their older peers who are not Luddites are also going to do great. But an 18-year-old—or, by the way, a 24-year-old, or a 14-year-old—with AI: we are going to see superproducers the likes of which we’ve never seen in the world.

So, yes, by the way, this is going to greatly stress the child labor laws.

>> [laughter]

Yeah, exactly.

Let me just say: the children yearn for the AI minds.

Yeah.

Speaking of that, we talked about Zoomers in previous episodes and why you like them: they have so much courage and are willing—or are sort of fed up—because they grew up in COVID school and all these adjacent impositions.

One thing you quoted me on recently was Chris Arnade’s post about how people talk about the educational divide, but there’s also a generational divide. Boomers are much more confident in their truth, while younger people are more post-truth, relativistic, and pluralist. I thought that was a really interesting epistemological divide. What did you find interesting about it, or how do you see that play out?

There are really 2 parts to it, which is very interesting. Part 1 is that a lot of boomers—somebody once said the definition of a baby boomer is somebody who believes what’s on the TV set. They believe what the talking head on the TV says.

Anybody who’s 20 knows that you obviously don’t do that, right? That would be stupid. But every 60-year-old or 80-year-old has been watching TV their entire lives, and when they grew up, it’s the old story we’ve all heard a million times: Walter Cronkite used to tell us what the truth was. Of course, that was always bullshit, but nevertheless, that was what the boomers believed. They believed what the TV said. They believed what The New York Times wrote. They believed these things.

Anybody below the age of 40 at this point just has example after example after example of how obviously that’s not true. Anybody who’s 20, who’s been through the last 15 years in school, just obviously knows that these people are fake, that this is not real, and that you just can’t take this stuff seriously.

Part of it is that divide. The boomers had—there’s this great YouTube account with an amazing video on this. There’s a great YouTube account called Academic Agent. It’s a British author named Nima Parvini, who writes these really interesting books. He has this 2-hour video that’s really worth watching called “Boomer Truth.” It’s a 2-hour documentary on this concept he calls “Boomer Truth,” which is basically, “Whatever the TV says.”

So, there’s the Boomer Truth thing. But then there’s this other thing, which is a key part of Boomer Truth: there’s no fixed morality. You get to make up your own values. All cultures are the same—moral relativism. All cultures are the same. Western society is not superior. There are many different cultures, and they’re all wonderful. It’s all great, it’s all great, it’s all great. If anything, the West is the worst of the cultures; the other ones are better.

Before there was “woke,” there was political correctness. The political correctness when I was in college was literally around what was called multiculturalism, or “multi-culti.”

Peter Thiel and David Sacks wrote about it in their 1995 book, “The Diversity Myth.”

“The Diversity Myth.” It was actually a term called “multi-culti”—multicultural. There were furious debates. There’s a classic book from that era—Peter’s book is great, but before that, there was a famous book that was huge headline news all through the country when it came out called “The Closing of the American Mind.”

Yeah.

It was this right-wing academic at the University of Chicago who basically said, “These colleges are teaching these kids that there is no morality. It’s all just morality is choose-your-own-adventure.”

There is this moral relativism at the heart of Boomer Truth. It’s this weird thing where there is a fixed, received belief that there is no fixed morality. The entire media apparatus, the entire cultural program, and the entire educational system got designed around this. All of the crazy stuff that kids are getting in school now is downstream of this movement from 30, 40, or 60 years ago.

If you’re 20, you’ve just come up in this weird environment in which, on the one hand, you’re like, “The boomers have no credibility at all, because I can’t believe they still believe what’s on TV.” On the other hand, to the extent that we do listen to anything they say, they keep telling us not to judge anybody and not to judge anything, and that all moralities are equal and all cultures are equal.

Of course their successors are going to come out of that with an incredible level of skepticism. By the way, this is not an abstract exercise, because these are the kids who came up through COVID. These are the kids who came up through woke and through all of the craziness of the last decade, the last 15 years.

I think these kids are coming out with a completely different viewpoint on how the world works. Not in every case, but in many cases, it’s completely different—much more, I would say, almost simultaneously open-minded and critical; much more interested in ideas; much more skeptical of authority and received wisdom; and much more cynical about manipulation.

They’re also much more sensitive to the media environment. They’re much more aware of the idea that there actually is psychological warfare going on, and they have been on the receiving end of it. They’re much more skeptical of authority. Their view of the authority figures they’ve seen in their lives, in many cases, is complete contempt—and in many cases, very well earned.

So, yeah, it’s a starkly different worldview than the boomers had.

Also very different from my generation, Gen X, and also very different from millennials. It’s something new, and I’m very excited. I think they’re fantastic.

Speaking of something new, would it be fair to summarize maxing as Stoicism meets “you can just do things”?

No, I think it’s just “you can just do things.”

Okay. [Laughter]

I think I can see what you’re driving at, and I think you could probably explain it that way, but I don’t know. The way I put it is, the Stoics put a lot of time and effort into trying to be stoic. [Laughter]

Whereas the whole point of maxing is that you’re not supposed to put that level of time and effort into being the way that you are. You’re just supposed to do it. So, yeah, I guess you could say our friend Ryan Holiday is a Stoic and not a maxer, as he demonstrated this week. [Laughter] Maybe right there in that video, you can see the difference.

Yeah, yeah, that’s well said. Last question from the chat, and we’ll get you out of here. How are you such a good monitor? What is your secret to monitoring so many situations? Any strategies? What is your approach?

Well, of course, being plugged into the MTS fire hose is absolutely critical. And, of course, the amazing tools that the team is developing and putting online are fantastic. I’ve been, among other things, glued to the coverage of the OpenAI trial this week on MTS. Is it MTS? Is it MTS.com?

Okay, yeah. Yeah.

Yeah, for sure. I long ago plugged the back of my skull—I wire-jacked into social media. So I have my continuous X feed, my continuous Substack feed, and my continuous YouTube feed. And then, as usual, I try to read enough old books to have some kind of relevance to the daily fire hose.

Yeah. Awesome. Well, Marc, thank you so much for coming on. Another great episode at a16z, and we’ll see you back soon.

See you soon.

Okay.