H-1B Shakeup, Kimmel Apology, Autism Causes, California Hate Speech Law

We’re back. We got the original crew here. It’s a tight foursome. With me again, he’s returned from, I believe, the UAE and MENA: the one, the only, your chairman, your dictator, Chamath Palihapitiya. He puts the dick in dictator. That’s what they all say. How you doing?

Good. You?

So you went and you got that—what’s that beautiful airline that we all take to the region?

Emirates first class? Yeah.

Oh, Emirates first class with the cabin.

It’s insane.

With the wine. Take everybody in, because I do business class for $1,400 round trip.

Emirates is unbelievable. But the problem is there are literally 1,000 movies. 1,000. So you have to favorite 30 or 40 of them. There were 95 different menu choices. I had probably 8,000 calories.

Oh, really? And that was just the wine, I take it. Yeah.

By the way, the wine is incredible. The wine list had, like, 1996 Montrose, and I was like, “Is this an airline?” I had never seen an airline wine list. It was pretty strong. It was great.

Did you bring your sommelier, Josh? Was he in the cabin next to you with his own—

I didn’t need it. No, I didn’t need it.

I like that. I like that. And, of course, your Sultan of Science, you got a great docket for him. It’s kind of like the Super Bowl for Sultan fans this week—

Is it?

Because—

I hope I deliver, bro. You’re putting the pressure on now.

Well, Trump—

You’re putting the pressure on now.

—cured autism, and you’re going to comment on it. So this is a pretty big deal. It’s a pretty big deal.

I’m going all in.

You let your winners ride.

Rain Man David Sacks.

I’m going all in.

And as I said—

We open-sourced it to the fans, and they’ve just gone crazy with it.

WSJ’s queen of quinoa.

I’m going all in.

And then, of course, the one, the only—he puts the bazaar in czar. David Sacks calling in from his—

His Sultan—

—his bazaar czar.

You got that backwards. He put the czar in bazaar, not the bazaar in czar.

Yeah, I was playing with it. I’m workshopping it, so I’ve been trying both ways. But yes, you’re calling in. You’re echoing, which is great.

I’m on the road.

You’re on the road between meetings?

Yeah, basically. Exactly.

So are you in a motorcade?

No. No.

Okay. Listen, the topic of the week: H-1B visas are being overhauled. The Trump administration announced a new $100,000 fee for all future H-1B applications. It’s a one-time fee. There’s been a little confusion about the details, but that’s how they do things in the 47th: some excitement, a big announcement, and then we figure out the details. Lutnick originally said it would be $100,000 a year, but then the White House clarified it would be a one-time fee. This is a huge jump.

The current fee is nothing. It’s like $2,000 to $5,000 that you pay to the government. You might pay a lawyer double that or triple that to do the work for you if you’re a big corporation. But this hits on a lot of the Trump campaign promises: tougher on immigration and looking out for U.S. workers. We’ve talked before here about the abuse in the H-1B system. I’ll give some of my personal insights on that after I throw to you, Chamath.

Before I do, there was an interesting Polymarket: “Will courts block Trump’s $100,000 H-1B by September 30?” There’s a 3% chance of that happening, so it looks like everybody’s kind of aligned with this program. All right, Sacks, I know you’re on the road, but your fans demand to hear your take on this. What’s your take?

I think it’s a good idea to have this $100,000 fee, and I’ll tell you why. Right now, there are something like 5 times as many H-1B applications as there are slots. I think they grant about 85,000 H-1Bs a year, and many more apply for them. As a result, you have a lottery where they randomly choose who the winners are going to be.

If you look over the past decade, roughly half the H-1Bs go to IT consulting firms.

Yes.

The average salary is about $65,000 a year. So it puts the lie to this idea that you hear that H-1Bs are for high-skilled engineers, AI researchers, and things like that. That’s not what happens in practice. In practice, you have this lottery, and a huge chunk of them end up going to low-end IT jobs.

I think by putting this $100,000 fee on it, you encourage the applications to go to the actual higher-skilled, higher-paid jobs where there’s actually a shortage of Americans, and you encourage U.S. companies to try to fill those jobs with Americans first. Putting aside some of the details, I think the big picture here is that they’re using market forces to put some scarcity around the H-1B application.

I think what that’s going to do is encourage applicants to actually be these higher-paid, higher-skilled jobs that the program is supposed to be for, instead of these lower-end IT shops.

Yeah, these are supposed to be for highly specialized workers. I can tell you, when I was in IT in the early ’90s, the abuse was happening all the time, and it was indentured servitude. It was disgraceful. The IT people would hire, typically, Indians, and they would say stuff to the effect of, “These guys are going to work for half as much and twice as long, and they can’t say no. That’s the best part. They can’t say no when we put them on weekend coverage. They can’t say no if we want to do a build-out and they have to work 10 days in a row, because we can kick them out of the country, and they have 30 days to find a new job.”

And so it’s a giant scam on the bottom half of these. I witnessed it firsthand. Every discussion I’ve ever had about H-1Bs in relation to IT and consulting has always been about saving money. The truth is, it’s been abused.

I talked about this in 2015 on CNBC, Chamath, when Trump first started to talk about it. He’s been on this for a while, and it’s just great to see them—

I had suggested $20,000 a year, and that’s kind of where they wound up. I additionally think they should do an auction for one-third of these. Let all these big tech companies that are truly trying to get in very unique PhDs from Oxford in AI put out how many they want to buy and at what price.

Do a reverse auction and fill one-third of them with—I don’t know—maybe OpenAI, xAI, or Microsoft. They jump the fence and pay $100,000 or $200,000 per person. What do you think, Chamath, broadly speaking, on this—the policy, the abuse, everything?

I came to the United States initially on a TN visa, which is the NAFTA visa between Canada and America. Then I switched to an H-1B, and I got my green card and my citizenship in the early 2010s. Elon came in on an H-1B. Sundar Pichai came in on an H-1B. Satya Nadella came in on an H-1B. There are a lot of folks who have done a lot of good things who have used this specific visa.

That being said, I think Sacks is right that people have found an end run and have been abusing this H-1B system. There was an incredibly exhaustive thread by Robert Sterling, I think it was about a year ago, but I wanted to use that as a jumping-off point to explain a couple of reasons why I think there’s been rampant abuse.

The first thing is that the H-1B program is supposed to be 85,000 visas a year. But here is the data: in many years, including the last several, it’s been upwards of 10 times that number. There are a lot of people who are getting shoehorned into this program. When you see this, you can start to see why a lot of people are saying that there is wage suppression and that it’s taking away from American jobs.

If the program was meant to be for 85,000, you would think, “Well, listen, that’s a drop in the bucket. Nobody would feel that in the American economy.” But when you start talking about almost 1 million people a year—600,000 to 1 million a year—that starts to be perceptible. That is absorbing a lot of revenue and wages that would otherwise go to domestic-born workers and legal immigrants who are already here.

That’s thing number 1. Thing number 2 is that there was a myth that H-1Bs were extremely highly skilled people. What Robert found in the data is that, actually, no, that’s not really the case. I think the average salary—I just want to get this exactly right—is slightly under $120,000.

If you started to tell me that these were the best-in-class PhDs in all of these whiz-bang industries where the companies are raising billions and billions of dollars, you guys already know that this salary would not pass the smell test. Most executive assistants at tech startups make more than $119,000 a year.

The idea that some qualified grad is making this should already set off alarm bells that maybe where there’s smoke, where there’s fire. So that’s the second thing. Number 1, we’ve been overallocating by 5 to 10 times. Number 2, these salaries aren’t the incredible salaries that you think of, which tends to mean that there is the potential, as Jason said, for some form of indentured servitude and wage suppression.

That's not good. And then the third thing is, you would ask the question: Who gets these things? It turns out, as Sacks said, a large plurality of these visas don't actually go to American companies that are looking to hire talent to make this American business do better. These are foreign companies that are arbitraging labor and bringing people in.

So crazy.

Cognizant is not an American business.

Tata.

Infosys is not an American business. Tata, Wipro. It's not to say that that in and of itself is wrong, but you need to find the right visa class to do this under.

Yeah, they're hacking it.

And so when you put all of this together, I think the broad takeaway is that, from where this started and what it was intended to do, we've deviated pretty wildly, and I think that this is a very important reset.

Now, the last comment I want to make is about the people that say, "Hold on, we are going to cut our nose off to spite our face, and it's going to stop an inflow of incredible talent." What I would just remind people is that it is really important to remember that when you are in the United States for a master's or PhD, you already get an automatic visa. It's called OPT. So you have multiple years when you graduate from a useful degree program in the United States to find a job.

I have several of these folks who work for me at $80K, $90K. These are incredible grads from Carnegie Mellon. They are off the charts smart. But because they did a master's or a PhD, they come with a couple of years, and you can oftentimes extend that, and that will give us a very good amount of time to figure out how exceptional they are. Then, quite honestly, I would gladly pay the $100,000 to get these guys on an H-1B program.

So I think if we're going to try to return this to what it was meant to be, which is to help American companies excel and get the best and the brightest, these changes are very good measures to course-correct and get us towards that.

1. America Needs Operation Paperclip

And just to give people the history of this, it was something that was started after World War II to get really specialized people, like Polish and German geniuses, building out rockets. I had this gentleman, Colin, on This Week in Startups, my other podcast. He applied for a product manager position and sent a résumé with the requisition number for the specific job at New Relic that he wanted, Friedberg.

He did this because, in order to have an H-1B visa, you have to put the job in a newspaper, right? So what these companies allegedly are doing is putting these jobs in obscure newspapers so that Americans don't see them. They're not putting them in places that you might see them. And there's a group of Americans who are going and finding these jobs and saying to Americans, "Go ahead and apply. Here's the shadow jobs," I think is what they call them.

He put the reference number in there, and they wouldn't even interview him. I talked to him, and he's crestfallen. He's like, "I would like to apply for this job, but it's obvious that I can't get into it." I think this just shows the entitlement of these tech companies, and I don't know New Relic's position on this. They can email us, and I'll give it in the next episode.

But they're basically listing fake jobs. And somebody, Abby, from People Ops over there just doesn't even let him interview for the job. The whole thing is just really dirty at the low end. And at the high end, it's under-monetized. So, Friedberg, your thoughts on this? I know you have a lot of friends. You're an immigrant yourself. I'm not sure how you got here and what visa you came under, but I think it came when you were a kid, right? I'm not sure what your parents came under, but what are your thoughts on this and the impact it might have?

Well, I think there should be 2 separate programs for what we could call highly skilled workers. What you were referring to: At the end of World War II, there was a secret US operation called Operation Paperclip, where we tried to recruit German scientists and engineers. Between 1945 and 1959, America recruited, I think, 1,600 of these scientists.

So it was both disabling to an American rival or adversary, but also expansive, because that was when the nuclear industry was growing and much of nuclear science was being pioneered in the earlier days in Germany. So the American workforce expanded, but more importantly, a new industry was enabled, unlocked, and grown in the US, and the German state was disabled by losing these scientists.

One could make the case that a similar sort of scenario should exist today, that we should have a second Operation Paperclip. Perhaps it should be a continuing process rather than necessarily this laissez-faire process that we have today, where we identify some of the top industries and the top scientists in top domains and go after those scientists proactively with government action, government support, in partnership with private industry.

If you look at papers being published across mainstream scientific journals, the majority of papers today across nearly every scientific domain are being published out of China. And this ranges from physics to chemistry to materials science to biotech, and there's a real case to be made that perhaps those scientists would be better off, and America would be better off, if they were doing their research, pioneering here rather than there.

I think there's a very good strategic case to be made that perhaps a more directed, high-energy, high-effort kind of Operation Paperclip be undertaken again around the world. The H-1B program, I do agree, has been heavily abused as a way of compensation arbitrage.

If you find highly qualified, excellent talent, as we all know, for a highly skilled laborer in engineering or science today, that person—if you amortize the H-1B over 7 years at $100K, that application fee is $15K a year, call it—certainly seems worth it for the right sort of talent, and it forces the question: Can this person be found in the United States or not?

The alternative would be to force a higher salary range such that you as a company are now basically being forced to pay a higher salary, which means you have to justify that this person is worth it to bring them in from outside the US, and you can't find the talent locally.

I'll tell you a program where we do this where it doesn't work in the US: It's called the H-2A program. This is the immigrant farm worker program that we use for temporary labor on farms. And the way that program is set up today is, you have to pay the farm worker that comes in on an H-2A some amount over minimum wage, and the amount that you have to pay over minimum wage is a function of the average wage in that state across all industries.

In the case of Florida, they're paying $5 to $10 over minimum wage for farm workers, and they cannot get any Americans to work on the farm. And they're being forced to pay $5 to $10 overage. By the way, these farmers and these farm businesses are being heavily subsidized by the government.

One way to think about the ridiculousness of what's going on is that the US taxpayer is paying a premium salary to foreign workers. What we should be doing is enabling a free flow of labor, but only in the case where there are no workers available in the US. But there is a downside to that model, as we're now seeing in the ag industry.

Farmers are losing money across the board, and they're having to pay a premium for foreign workers to come and work on the farm, and they can't get US workers. There's 2 sides to the sword on this, is my point, but I do think this Operation Paperclip notion should be taken on as a separate kind of strategic mandate.

Absolutely, yeah. And a lot of the Jewish scientists had already fled Germany.

That's right.

My understanding is that Paperclip was for the Nazis, the former Nazi scientists.

That's right.

And they were working on some pretty dark and

cutting-edge stuff in chemical and biological.

Well, it was everything.

It was everything.

Remember, at this era we were just developing quantum theory, and quantum theory led to nuclear science, which led to the development of the atomic bomb. So, yeah, I think Operation Paperclip was pretty far-reaching.

But today, as an American, do you really want all of the cutting-edge research in materials science, in physics, in chemistry, et cetera, to accrue to China? Or should we be thoughtful about it? It's intellectual talent that's making these breakthroughs. It's not necessarily institutional capacity. It's not like they have better institutions per se than we do. We have amazing institutions, amazing capacity, amazing places to live, and so on.

So there's a real mandate that we should probably think about undertaking here, not just for extension of our industrial base, but also for disabling what we would consider rivals or what America might consider rivals.

Just on this Operation Paperclip point, it's interesting that the Chinese government took away the passports of the engineers at DeepSeek after the launch of that model, or at least it was publicly reported. I can't attest to this from firsthand knowledge, but there were definitely a lot of reports about this, and you can see why.

If we could recruit or snap up a few hundred or, at most, a couple of thousand of these top AI engineers, that'd be a game changer—

Totally.

—in the AI race. So China is starting to see those people as a strategic asset, and they're not going to let them immigrate, I don't think. But it would be something for us to think about. Certainly on the chip design front, there's probably just a few thousand people—that's all we're talking about—that would be a game changer on that side, although they're largely in Taiwan, not China. And again, I don't think the Taiwanese government is going to be too excited for us to snap them all up and move them to America. But in both these fields, there are a relatively small number of people, kind of like in the space race, who, if they were all in America, it'd be a huge game changer.

Here's the thing. We do have a really rich diversity of people from all around the world in higher education institutions in the United States, getting master's degrees and PhDs. We just need to be better organized about what to do with them, and we need to reach out to those people, build relationships with them, take advantage of OPT, and then we can always create a different class of visa for them. We have the ability to do these things called national interest waivers.

So all of the infrastructure exists, and I think that if we can clean the decks on the H-1B stuff, it'll give people a lot more incentive to support the national interest waiver concept. I think the reason why people don't believe in this entire immigration conversation is that, on every part of the distribution of immigration, people see problems. They see an open—

Abuse.

—border on the one side.

Well, they're worried it's going to be abused. Yeah, exactly.

Exactly.

Yeah, it's abuse. And it feels terrible.

There have been too many of these horror stories where an American is told to train their H-1B replacement who's making 25% as much.

Look, I shouldn't even tell this story, but it was told to me yesterday. I don't know this to be true, so I don't want to fan the flames of speculation here. But what I was told is that, in certain countries, there are these puppy farms that essentially get these kids onboarded into U.S. colleges. They're not the great colleges, but they are decent enough, in all the far-flung corners of America. They get them into master's programs, they pay for their school, and then these folks have to send money back to pay off their degree. So if that's happening, then these kids are being abused as well, right?

Yeah.

So the whole thing has just completely run amok, and I think we need to clean it up. At that point, we have the chance of rebuilding trust, where then we can propose what Friedberg talked about, and everybody would be supportive because they see that the system works as intended.

I think closing the border has made people feel a lot better about it, because that was so abused that people just look at immigration in one bucket and they don't separate it into multiple buckets. There's compassionate people who are true dissidents, who we want to show compassion for. It's a small number of people. Then you have all these folks who were being taken with mules and coyotes over the border, then using that same compassionate designation and abusing that.

When you do see that abuse, I think people, when they hear Trump come on here, and when he was a presidential candidate, President Trump came on and said, "Hey, we're going to staple a green card right to those degrees," and he immediately got backlash. Well, now I think, if things have calmed down and President Trump and the administration have the high ground, they can say, "Look, we closed the border. Now we should have a really thoughtful discussion. We're giving these people American educations. They want to start companies. They want to build our companies. They want to build products and services. Those products and services are going to create more jobs. And look, we're at 4% unemployment, 4.X. We're in good shape because we closed the border, and that's where you had millions of people coming in. That was the true problem, and this isn't the actual problem."

By the way, Sacks, you make a really great point, Jason. Using the one word “immigration” to wrap up all the people who are coming to the United States, I think, masks the real series of things, or set of things, that are underlying: people seeking asylum, people in need, but then also people that we want to actively attract and bring here. It's probably not the right term to use, just the word “immigration.”

Yeah. The word you want to use is “recruitment.”

But it should probably have a qualifier for every term.

Yeah.

Yeah.

Yeah, the term is “recruitment.” What you described, the Paperclip model, that's recruitment. Then you have dissidents, and that is compassion for true dissidents. And then there's—

And family.

—and family. Great. So you have compassionate dissidents and family members. Then you have this big thing in the middle, which is everybody in the world wants to live here. Everybody can't. That's immigration. So just put it in 3 buckets, and then our leaders need to have discussions—

—three different discussions and have it in a thoughtful way, not muddy the waters and politicize this. That's what's been causing such a big problem. Both sides of this argument have been so charged. Hopefully now that the biggest one has been deftly done by Trump.

Trump said it at the UNGA. He said, "You know how many people have crossed the border since I came into office?" He said, "Exactly 0."

Yeah.

That's incredible.

I mean, we have the greatest military in the world. We can't close the border to people who are coming with backpacks across a border? We should be able to secure the border with drones and cameras.

It was done deliberately. They opened the border, and you know that.

Of course.

Because remember when Texas tried to put up barbed wire—

Yeah, they were—

—to enforce their own border.

That's right.

The Biden-controlled Border Patrol got rid of it. They removed it.

That was a tell.

So this was done deliberately.

Yeah. That was a perfect tell. I remember when we covered that last year. That was the total tell.

2. The Autism Causes Debate

Okay, Friedberg, some major news in science this week. Let's talk about autism and the press conference that happened this week with Bobby Kennedy and President Trump. Here is the chart. Autism has increased dramatically over the years. There's a big debate about what's causing this, and there are obviously correlations and causations. There's the testing of this, and maybe we're just testing a little too loosely around this. We went from 1 in 10,000 in 1970 to 1 in 1,000 in 1995 to 1 in 32 in 2022.

The press conference, Friedberg, was a little spicy and unique. “Performative,” maybe, were some of the criticisms. But there's a real issue here, and why don't you take us through it and educate us so we can get to reality? Because the press is having a field day with this, obviously, on both sides.

I think autism, just like Alzheimer's, may have several underlying conditions that lead to what we would call the phenotype of autism. That is what we all observe as autism. It's considered a spectrum disorder. There are many different variations of it. There may actually be many different underlying conditions or biological drivers that are causing it.

One of the drivers that came up during the press conference and in the subsequent interviews Marty Makary, head of the FDA, has done, is that they've identified and shared papers that have been out for some time showing that there is a receptor that absorbs folate, a type of vitamin B, and that that folate receptor may be attacked by the immune system. As a result, you can't really take up vitamin B. Those cells become dysfunctional, and when those cells are dysfunctional, you end up having what looks like what we call autism.

One of the things that they announced is they're going to work on getting the label updated for leucovorin, which will resolve, for many people, the folate receptor issue. The other thing they brought up is a paper that was done by Andrea Baccarelli, who's dean of the Harvard T.H. Chan School of Public Health. This paper is a bit old. He took several studies and analyzed them and showed that, across 46 studies, 9 of them showed no association with acetaminophen, the main active ingredient in Tylenol. 4 showed a negative association, meaning it was actually protective and good for the fetus, and 27 had a slightly positive association, which means that it was having some contributory effect on both ADHD and autism spectrum disorder when women would take acetaminophen while pregnant.

Nick, if you want to just pull up that image from the paper. This is the original paper that was published by Baccarelli. Again, he didn't do any primary research. He didn't actually go and study patients. He took the data from 46 other studies, and then he added it all together to run this meta-analysis. You can see here that he showed some risk. There's no specific way to quantify that risk, but there's some increased risk of having attention-deficit/hyperactivity disorder or autism as a result of taking acetaminophen while pregnant. Now, I think autism, again, one of the underlying conditions might be this autoimmune condition associated with the folate receptor.

What causes autoimmunity is a whole ’nother conversation, and we can get into the vaccine stuff if you guys want to, because there are obviously a lot of conversations going on right now about the immune system being primed to have an autoantibody response. But there may be other things contributing to it.

So I think it’s pretty clear that, in our modern world over the last couple of decades, there’s a cumulative effect of environmental exposures that children are getting, whether it’s microplastics, whether it’s chemicals in the food, whether it’s just the environmental exposure in the air related to small molecules, or whether it’s related to other things we’re putting in our body. Every one of these things—the way to think about it is, maybe if it has a positive effect, it might increase your chance of autism by 0.05%, and then another thing might increase your chance by 0.07%, and so on and so forth. And so when you add up all the things in our environment, there may be a cumulative effect that results in different underlying conditions in our body, which may result in what looks like things that we call autism. And so none of these are very specific. There’s no one shot and one path—

Okay.

—and one specific thing. I think that’s very important.

Yeah. Let me just ask 2 clarifying questions, really lightning-round, for you, Friedberg. Number 1, for the audience, how is autism diagnosed in these studies? Is there a blood test, a genetic test, or is it just a bunch of questions? I know the answer, but I wanted you to clarify it for everybody.

And then how do the geographical differences in autism—just like we saw with trans kids, where there are many in certain cities and none in others—fit into this? Maybe you could talk a little bit about those 2 issues, which I think many people—

Yeah.

—have been talking about.

So I don’t think that there’s 1 specific diagnostic test for autism, as if it were 1 disease. Again, these are phenotypes, these are behaviors that are being measured that people call autism spectrum disorder. And so the diagnostic criteria fall under a set of screening and behavioral tests that go on. One of the things that—

A survey.

There’s a survey, there’s an observation—

It’s a survey, yeah.

—and there’s a bunch of things that can add up to make a—

Is it like a score? Is it a score?

Yeah, so there’s a scoring system, exactly. And so then there are different levels.

What did you and Sacks score?

Yeah, Maurice.

Because we could bet. Hold on, before you tell us, Chamath and I are going to bet on it. I’m going to say that Sacks beat Friedberg by 16 points.

Well, then make it over-under. Make it over-under.

Over-under? I think these guys are—

Well, who’s highest on the spectrum?

Well, Sacks may have gotten in the 90th percentile.

He’s definitely in the 90th percentile. I put Friedberg in the 70th percentile.

Oh, great.

We need a Polymarket. We need a Polymarket. What did Sacks and Friedberg get on the score?

Polymarket.

3. The Tylenol Evidence Fight

By the way, just to go back to the acetaminophen study—

It’s so great.

Oh, look, there’s the emotional detachment. Go ahead, Friedberg. Go ahead.

Yeah, just no reaction. You and I are laughing.

I’m making an active choice not to engage.

I am making an active choice to repress my emotions.

One thing to note, and one of the controversies about all of this, is that the paper published by this guy, Andrea Baccarelli, from Harvard—I don’t know if I’m pronouncing his name right—has actually been challenged. In 2023, he was called as an expert witness in a lawsuit against the maker of Tylenol, and in that lawsuit, the judge threw out his testimony as unreliable because he was being paid $150,000 to give the expert testimony and work on the case. And so, because of the payment that he—

Wait, whose side was he on?

So this is the guy who published the paper that linked—

Right.

—that showed an increase in—

In the expert testimony, was he pro-Tylenol or anti?

Anti-Tylenol. So he was an expert witness for the lawyers—

Ah.

—that were filing the claim against Tylenol.

Got it.

I will say that, in the last couple of days since this press conference, he has publicly said that we are not yet certain or sure about the link. And I want to remind everyone that this association study indicates an increased risk, but that doesn’t mean that if you take Tylenol, your child is going to have autism.

Right.

That doesn’t mean that there’s determinism here. There is a statistical chance that there is a slight increase, and the more acetaminophen you take—

Well, no, no, no. Hold on. Hold on.

—the higher the chance, and that’s what the paper showed.

Let me take the other side of this. Here’s what we know. We know for sure that there is this potential autoimmune issue that you, as the mother, can express as an antibody. There’s an antibody test for it. We also know that the child can be tested for it. So at a minimum, we’re now at a point where we can create a very thorough, well-funded study to get to the core of this issue.

Separately, to the extent that you do test positive, there will be some doctors and some parents that may decide to take Leucovorin prophylactically, and then also to administer it to the child before it’s clear whether they do or do not have autism. I have several friends who have kids on the spectrum. I talk to them about Leucovorin. What they say is that when you have extreme autism, the drug is very effective, but when you have kids that are more mild on the spectrum, then there’s a lot of benefit from behavioral modifications and behavioral training, and it’s not clear how effective that drug is. But in 1 specific case, 1 of my friends is considering it for their child.

Here’s the point, though. I think the point is that we need to test for this, and I think that some combination of governments and industry should come together. Beyond that, the point that I want to make, though, is—and I think, Friedberg, I don’t think you’re adding to the conversation when you say that there’s no determinism, because on the off chance that there is, I would say that we don’t know yet—when you see women taking acetaminophen in this performative-art way, basically to try to pwn Donald Trump, and it’s all over TikTok and X, I just think it’s reckless, and I think it solves nothing. You’re supposed to go talk to your doctor about it under all circumstances anyway, but then when you transform it into some sort of protest vote without really knowing, I think that’s really dumb.

Yeah, and this test, by the way, Chamath—I’ll say this: folate receptor autoantibody—there has been a test available, I think, since 2012. This has been around for some time. In fact, there was a paper—

Can’t your patients take this test? Of course, because you—

Yeah, and there was a paper published on this. I’m trying to find this paper. I’m not sure how much follow-up there’s been. This was a SUNY paper where these guys went and found folks who scored very high on the autism spectrum disorder diagnostic test, and they found a high prevalence: over 70% had this folate receptor autoantibody. This was 1 paper, so I don’t want to give it a ton of credence. There’s been a lot of follow-up since then, and I’m not an expert in this space, but I did some research on what the history is of this.

And so this is a very well-known kind of correlative effect, and it could be a very big contributor, but obviously, no one should feel like, “Hey, if I don’t have the autoantibody, then my kid is fine.” That doesn’t necessarily mean that’s the case. Again, there may be several paths to autism, and the thing about acetaminophen is, just like, hey, have a drink or smoke a cigarette, the more you do, the higher the risk.

Right.

This is the case with anything we put in our body, and I think that’s the point about what they’ve identified in the paper that was published regarding acetaminophen. And so there’s still a lot to be determined about how they’re going to provide guidance to pregnant women: Do you have a fever? Do you take it for pain? What’s the right criteria? All that sort of stuff.

Well, let’s just say we all agree that medicating yourself to make a political point—

It’s just hilarious. What are we doing? That can’t be true.

It’s absolutely—

It’s stupid.

—it’s absolutely idiotic.

It’s idiotic.

Absolutely, whether this is true or this is a form of art—

It’s stupid.

It’s stupid.

Yeah.

Yeah.

I think for his next trick, President Trump should warn people not to snort rat poison.

Yes. Also not a good idea.

Somebody was tweeting, “Next, President Trump’s going to recommend not using toasters while you’re in the bathtub.”

Yeah.

My fellow Americans—

It’s this guy.

Donald Trump

In light of recent studies, I wanted to warn against the use of toasters in your bathtub—

Toasters.

Donald Trump

—or your shower.

Don't do it. Don't put the toaster in the bathtub.

You know people are going to do it if he tells them not to.

Exactly. I have such an inappropriate joke right now, but I'm not going to say it.

Let's take Friedberg's math. When I saw that—

Yeah.

—and you think maybe there's a 0.05% chance, well, guess what? If you stack that up and it actually turns out to be true across the thousands of women who then performatively did this idiotically, you could actually have an extra kid or 2 with autism that didn't need to have it. What is going on in America?

Well, look, I think there are a lot of people who instantly had a snarky reaction because it was Trump and Bobby Kennedy making these claims about Tylenol, and the media played into that. I don't know what the truth of it is, but there are—

Even Tylenol tells you not to take it. They say, “Call your doctor.”

Yeah, exactly. There are—

How stupid are these people?

There are plenty of articles establishing this risk. There's a paper from Johns Hopkins from 2019 called “Taking Tylenol During Pregnancy Associated With Elevated Risk for Autism and ADHD.”

Pull this up. This is, to your point, CNN's coverage when these acetaminophen studies came out historically. You can see that when it was announced at the Trump conference yesterday, they said, “Trump links autism to acetaminophen used during pregnancy despite decades of evidence it's safe.” If you read back, this is 2017 and 2016: “Acetaminophen during pregnancy may increase risk of hyperactivity. Studies link acetaminophen in pregnancy to ADHD.”

So when Trump says it—

So when Trump says it—

—they take the negative angle.

In the headline, they dismiss it. But every other time, they've taken the paper and published a news article on the paper itself. I just couldn't believe that CNN—

I mean, you just don't get news.

They've published a news article on the paper itself every other time.

They ignore all their previous reporting on the subject.

Right.

That's what's crazy about it.

This supercut was hilarious. Check out the supercut of Trump saying “Tylenol.” It was hilarious.

Donald Trump

Don't take Tylenol. Don't take it. With Tylenol, don't take it. Don't take it. Don't use Tylenol. Don't take Tylenol. Don't take Tylenol. Fight like hell not to take it. I think you shouldn't take it.

It's just hilarious how he presents information: “Just don't take Tylenol, don't take Tylenol.” Great week for Motrin, though. I heard sales are up for them.

Well, no, because it's also been established—

I know.

—for a very long time that you don't take ibuprofen or aspirin.

Yeah, you definitely don't take ibuprofen.

Those 2 things are even worse.

Yeah.

Yeah.

Even worse, yeah.

Ibuprofen is terrible if you're pregnant. Do not take it.

So wait a minute. Why is it specifically that this one mechanism of action was completely inoculated, and every other mechanism of action for things like pain relief and headaches is known to be pretty bad for women and babies? Come on, guys.

Yeah. Friedberg, explain to us the disparity in different regions of the world and autism. Would you think that's the surveys, and maybe people in the United States are a little more—

Again, I'm not an expert in this stuff, J. Cal. I can opine on what I've read, and I've done some studying. But think about my level of knowledge being in the range of hours compared to yours—

Oh, well, no—

Yeah, but I would say—

I'm a complete expert on this because I've watched all seasons of Love on the Spectrum.

Right.

And I love that show.

Yeah.

Have you seen Love on the Spectrum, Sacks?

What is it?

Oh, watch it with Nat. She'll love it. It's great.

I do agree with the general notion that we have a cumulative effect. Many studies that are done, by the way, are organized around short duration and high exposure when it relates to novel molecules in the environment and how we're exposed to them.

Yeah.

The cumulative, long-term effect of these things—as we learned recently with respect to the amount of microplastics and the endocrine-disrupting nature of microplastics, where they can actually bind to specific receptors and, as a result, block the expression or binding of other proteins, which can have a systemic effect on you—we didn't realize that until recently, after we've been making plastics for 70 years. Then we didn't realize these plastics were breaking down and cumulatively growing in our environment, in our water supply—

In our balls.

—as a result, in our food—

You even said—

—in our balls, in our brains, in our hearts. What are these molecules doing as they accumulate in our body? That now needs to be studied and understood, and then we have this big, challenging industrial question of what do we do about it. But that's true for nearly every novel molecule we've developed over the past 100 years on Earth. Some of them, by the way, we could make the case—we can show deterministically—they break down, they don't persist, they're fine.

And then some of them, we could say, look, there's a cumulative effect. As more of these molecules enter our body and some of them stay in our body, they over time increase the probability of things like DNA breaks resulting in cancer, or things like endocrine disruption resulting in metabolic effects, or things like what we might be seeing now with folate receptor autoantibodies that drive autism spectrum disorder. I think there's a lot to be said, generally speaking, about the cumulative effect of a lot of the stuff we put in our environment. As much as I could disagree with things that Bobby Kennedy might say, I can tell you asking these questions is critically important. As scientists, we have to constantly interrogate, and this cannot be just a political point. I'm not going to sit here and say I agree with the statements that might be made by the president when he says never take acetaminophen. I think it's a little more nuanced than that, frankly.

But I do think that the idea that we should be asking the questions and interrogating for answers is very, very important. And so I—

I think that's an important takeaway because if you talk to people who have been impacted by this and people in the community, the reaction was, “We're glad we're getting a lot of attention. Yeah, there was some weird press conference on the margins, but we're glad that this is a focus now.” I think people are also glad about vaccines. Again, I don't think we should be banning vaccines, but I do like the idea that we're questioning all these things and we're not just giving the medical-industrial complex a complete pass on this, and that we're spending money and investing in it. So let's go on to our next story around censorship.

How much money do you think is spent in the USA annually on autism?

Gosh.

Do you think it's enough to fund a longitudinal trial?

I don't have the data to know, and I guess on a societal basis, you would be thinking, “Is there a better use of capital?” But we're a rich country. We have the ability to fund some of this stuff, and there are private companies who—

Well, I bet you—

—probably would fund this too, right?

I bet we're wasting money in all kinds of random places where—

Yeah.

—finding a few billion dollars to run this trial is a good thing.

Yeah. That probably makes sense. All right, we've got a lot of news in the censorship space, so this is “This Week in Censorship.” So much to talk about. First story up: Jimmy Kimmel is back on the air. Earlier this week, Disney announced that ABC would resume airing Kimmel's show, and I think that was Tuesday night. Disney explained why they suspended Kimmel last week: to avoid further inflaming a tense situation at an emotional moment for our country, and they called Kimmel's comments ill-timed and thus insensitive.

He came back and had a massive amount of reach, but Nexstar and Sinclair's affiliates decided not to air it, and that was 60% of the market. Polymarket had Jimmy Kimmel canceled by September 30th. It was nearly 80% after Kimmel was suspended last week, but that's plummeted to 1% since then. It was quite emotional. I'm guessing everybody watched it—

No, I did not. I did not watch it.

Oh, you didn't watch it. Okay. We'll play a clip here. It was incredibly heartfelt—

You should watch what he said.

It was incredibly heartfelt and deft in terms of its execution. It was sincere, and here it is.

Jimmy Kimmel

I don't think what I have to say is gonna make much of a difference.

If you like me, you like me. If you don't, you don't. I have no illusions about changing anyone's mind, but I do want to make something clear because it's important to me as a human, and that is: You understand that it was never my intention to make light of the murder of a young man. I don't think there's anything funny about it.

I posted a message on Instagram on the day he was killed, sending love to his family and asking for compassion, and I meant it, and I still do. Nor was it my intention to blame any specific group for the actions of what was obviously a deeply disturbed individual. That was really the opposite of the point I was trying to make, but I understand that to some, that felt either ill-timed or unclear, or maybe both.

And for those who think I did point a finger, I get why you're upset. If the situation was reversed, there's a good chance I'd have felt the same way. I have many friends and family members on the other side who I love and remain close to, even though we don't agree on politics at all. I don't think the murderer who shot Charlie Kirk represents anyone. This was a sick person who believed violence was a solution, and it isn't. Ever.

Yeah, and he references later in this apology of sorts—I think it was pretty clear he was apologizing—his own faith, his Christianity, and just how beautiful it was that the widow of Charlie Kirk had forgiven the shooter, and he got broken up about that as well. He then went on to do a bunch of jokes and have a normal show. It had massive ratings. Obviously, everybody was tuned into it, and we'll see where it goes from here.

He didn't apologize. I didn't hear an apology.

Okay.

Did he apologize? Do you know what that means in the English language? My understanding, if you look at Twitter and stuff, was that there was not technically an apology.

Okay.

4. The Political Violence Debate

He didn't apologize, and he didn't say what he had done wrong. I guess what he said is, "I didn't mean to make light of—"

It wasn't my intention.

—the situation.

Yeah, it wasn't my intent. Yeah.

It wasn't my intent.

It wasn't his intent, but—

It was an explanation more than an apology? Okay.

Nobody was accusing him of making light of the murder. What he did, and what people were upset about, is that he lied and said that the shooter was MAGA, and he did not hit the nail on the head in terms of addressing that, and he's being called out for that.

Now look, I still think that his statement there, let's call it an apology, was constructive and positive because at least he is showing empathy toward the other side. He obviously feels bad for Erika Kirk and for Charlie Kirk, and in the current overheated political environment, just expressing empathy for the other side is a positive statement. I think he definitely brought the temperature down.

And I think later in the statement, he also makes an important point. He says, you know, "Just selfishly, I have threats on me," and what he's basically saying is, "Look, we don't want to get into a civil war here. We don't want to get into a cycle of tit-for-tat retaliation. Let's not play Hatfields and McCoys." Those are my words. He didn't say this, but that was sort of the intimation of what he was saying, and I think that is a good thing to say.

No one here should want a civil war, and this thing could go off the rails really badly. So look, I think that his statements were positive and welcome, and they showed empathy for the other side, but he did not fess up to what he really did wrong here, which was to claim that the shooter was MAGA. That was the thing that was deeply offensive.

Yeah.

And hold on, the reason why he did that is he was not the only one doing it. In the early days of this shooting, or this assassination, it was a talking point on the left that the shooter could be right-wing. And the reason why people on the left were saying that is it was exculpatory. It was basically to put the blame on the other side instead of looking in the mirror and copping to the fact that there is this rise of left-wing political violence and assassination culture.

As we demonstrated on the pod last week by looking at all the data and all the numbers, there really is this poisonous ideology that is on the left, and yes, there's some of it on the right, but way more of it on the left, that political violence can be used to solve problems. And the left really does need to look in the mirror and rid itself of that ideology. By not admitting that this assassin was motivated by that ideology, they are ignoring that opportunity for self-reflection and progress.

That last part is important. I think the point is that when you say that somebody is mentally deranged, what most normal people do is then say, "Oh, it was an aberration. It was an outlier." And I think that that is a dangerous way to try to sweep under the rug what is something that's more virulent and is increasingly acceptable in society.

This guy might have been crazy in the sense that he was willing to use murder to achieve his objectives. I think we can all say, on some level, that's crazy. It doesn't mean he wasn't animated by ideology that lots of people believe.

And I think the proof of this was the celebratory reaction to the assassination of Charlie Kirk. You saw it on TikTok, you saw it on Bluesky, you saw it in these corners of social media. You definitely saw it on Reddit, where you had thousands, maybe hundreds of thousands of people celebrating the death of Charlie Kirk and basically buying into this idea of political violence as a solution to their problem, and to the idea that it was acceptable to use violence against people they hated.

And so again, this is the problem with the random-nut theory: It really ignores all the evidence we have about the larger reaction to the Charlie Kirk assassination. And this is a thing that the left really doesn't want to confront. It does not want to look in the mirror here and say that we have a problem on the left with this assassination culture.

We talked about this last week. And if you look at polling, they just did polling around this, and there are still millions of people on the left who believe that the shooter was MAGA. Jimmy Kimmel helped foster that belief with this disinformation that he put out there. And he really should have hit the nail on the head in terms of saying that he got that wrong, that it was wrong to say that. He could have been a little bit clearer about that.

I'm not dismissing the positive things he said, because I do think it was good for him to show that—I mean, I'm going to give him credit for getting emotional. I know a lot of people on the right think that it wasn't sincere. I think that it probably was sincere. I'm going to give him credit for that. I think his comments were constructive, but he did not apologize for the thing he actually did wrong.

Mm.

And in fact, he just replaced that original lie with a new form of left-wing spin, which was the random-nut theory.

Okay.

And I think, let me stop there. But the point is that we really need to come to grips with the fact that there is this toxic political ideology now that's mostly on the left that does need to be confronted.

I'm going to go ahead and say we should clean up a little bit here, or I'm going to clean up. He should have made it clear that this wasn't a MAGA person, but I'm just going to repeat the quote: "We hit some new lows over the weekend with the MAGA gang desperately trying to characterize this kid who murdered Charlie Kirk as anything other than one of them, and doing everything they can to score political points from it."

Now, he should have said, "That turned out not to be true. It was actually a liberal." And I think it was not that he said the guy was MAGA; he said people were speculating he was MAGA. That actually was true. So just to be clear here, he never said the person was MAGA. He said—

Wait, who was speculating? The MAGA people were never speculating. Hold on. Charlie Kirk's fans, his loved ones, his coworkers weren't speculating about this.

No.

It was people on the left—hold on. People on the left, of which Jimmy Kimmel was one, were trying to plant the story that this assassin could be MAGA.

Well, this is the problem—

And that was done deliberately—hold on. That was done deliberately as disinformation to take the blame off of the political ideology—

Yeah, I don't think anybody—

—of the left and the culture of political assassination that's been rising on the left.

Okay, so over that weekend, there was definitive speculation that this might be a Groyper, one of Nick Fuentes's fans. There was also significant speculation, and this is why people shouldn't speculate in a breaking-news environment, because you'll frequently get it wrong. And people had gotten it wrong because they assumed that this individual was MAGA because his parents were. And then, when it turned out it wasn't true, that's when the record got corrected. But this was one of these instances where people actually were speculating, and they were wrong, and that's why you should always wait in a breaking-news environment.

Well, you're using a lot of passive tense in order to avoid who said what.

How so?

You said there was a lot of speculation.

Yeah, I'm talking about social media speculation.

There was an effort to create a narrative, a false narrative, that somehow MAGA was to blame for this. Jimmy Kimmel was one of the leading people who did that. We covered last week the rundown, the timeline of what was known at the time Jimmy Kimmel said that. And Megyn Kelly did an excellent job summarizing everything we knew.

We knew what was written on the bullet casings. We knew what the parents had said. We knew what the friends had said. Megyn's tweet on this is very exhaustive. So you're acting like there was a legitimate basis for Jimmy to say what he said, and there wasn't. And I don't know why you're covering for him right now.

No, no, no. I'm not covering for anybody. I'm clarifying a factual error. He did not say the person was MAGA. He said there was speculation that weekend, and that weekend there was speculation that this was a Groyper and that the family was MAGA.

By who?

On social media, everybody was speculating about this. And I think the problem here is that we're saying that the left wants to assassinate people. The left has assassination culture. I know many people on the left. I haven't talked to a single person on the left who is in favor of what happened or in any way supports it, and it's quite the opposite.

Every single person I know on the left, every single person who is a high-profile person on the left, with the exception of maybe 1 or 2 really dark people, has said that this is horrible and tragic, and there is no place in civil society for violence. So I'm trying to balance this out and say, this wasn't the left. This is not the left's strategy. The strategy does not want to murder people. That is absolutely false, and the left is not pro-assassination culture.

You certainly changed your tune, because last week we showed you plenty of data showing that there were 3 times as many people on the left who were celebrating or endorsing political violence. You didn't object to that then. You didn't show me data on the opposite.

Oh.

The only leader I saw was Ilhan Omar, who was saying, “Charlie Kirk's got a terrible legacy,” et cetera. Those were obviously inappropriate comments, but every other person uniformly condemned this. Now, I know there's tons of surveys out there, and I think if we're going to talk about political violence, the right also has a political violence problem that they need to address.

We saw on January 6th MAGA and all of these people beat police officers and destroy the Capitol. That is also political violence. It's not an assassination. It's obviously distinctly different, but they were beating cops, okay? And the price they paid for that was they were all given pardons for beating police officers, okay?

So there needs to be better leadership from the Democrats and the right on this, and everybody needs to calm this down and say, “This is not acceptable.” Whether you're beating cops up on January 6th, assassinating somebody, or threatening people that they have to fight, this is all terrible. Everybody has to calm everything down, and that's the example I'm trying to make here on this program: to have productive dialogue even though we disagree about things.

Every person I know who's on the left absolutely believes this is abhorrent, and they would never condone it, period, full stop. And every leader, with the exception of 1 or 2 people who—I don't understand why they would ever criticize a Christian who was murdered in cold blood—I'm sorry, that's totally unacceptable. But anyway, that's my position on it.

Look, I understand this is a heated issue. Let me just make one final point. I know that both sides have their quote-unquote nut cases, violent extremists who engage in horrific crimes, and they should all be denounced equally.

The difference here, I think, is that you saw thousands, maybe even upwards of 100,000 people on the left on social media rejoice and celebrate his assassination or downplay it and minimize it on the grounds that somehow he deserved it for the things he said. And I just have to say, I don't think we've seen that behavior before on the part of the right.

Whenever there's been some horrific crime, I don't remember anyone on the right ever celebrating that. It was not something that was mainstream discourse by any means. I certainly did not see thousands of videos and Reddit posts celebrating that. And I think what you see in the polling data is that, yes, there are some people on the right who feel that political violence is acceptable or a solution, but that number is 3 times greater on the left.

I'm glad JCal doesn't know any of those people. That's reassuring. But nonetheless, it's there in large amounts of data, and I think that we need to address that problem without minimizing it or both-sides-ing it, or else we're never going to make progress as a country.

5. AI Learns To Plan

Okay, JCal and Chamath both had to run. We started late today, and we ran a little bit too long for both of them. Sacks and I are going to wrap it up with a quick conversation on AI. Sacks, I don't know if you saw, but there were 2 papers that were published this week, each of which on its own I would say was pretty important.

Nick, if you could just pull this first one up. This is the MIT paper. This paper is called “Teaching LLMs to Plan,” and effectively what this team did—and again, they were out of MIT in collaboration with a scientist at Microsoft AI in Mountain View—is they basically created an instruction-tuning framework that teaches LLMs to do symbolic planning.

That basically means that the LLMs think step-by-step, or chain of thought, in a smarter way by making them generate explicit state-action-state chains. Then they trained the model by giving it feedback with an external plan validator, which is effectively going to be a human or a software tool that says, “Did this series of steps make sense to do the thing you're trying to do? If not, here's what you did wrong. Here's what you should have done better.” And they were able to achieve planning accuracy of up to 94% on some standardized benchmarks that are used for chain-of-thought reasoning and planning using LLMs.

This is a 66% absolute improvement over baseline models, and so this is pretty substantial. They took Llama 3, and they were able to increase the performance from 1% to 64%. The outcome of this basically is that this sort of system can be used to train LLMs to do better reasoning and better chain of thought in such a dramatic way that LLMs will look like they are starting to reason.

By effectively training them on the steps of planning and how to reason, the LLMs get better at looking like they're doing reasoning using this kind of symbolic planning method that they then built a tuning framework around. That sounded a little bit complicated, but I think ultimately what it translates to is they figured out a method to get AI to act in a more reasoned way in developing step-by-step plans and executing against those plans, and the results and the benchmarks are incredible.

So this was a big breakthrough, I would say, this week, Sacks. I don't know if you've spent any time looking at this paper from MIT or talked with your team about it.

I haven't seen it, but what exactly is the symbolic framework they're talking about? I understand chain of thought, but what is it that improves the accuracy?

There's an old language called PDDL, or Planning Domain Definition Language. PDDL is an attempt to standardize AI planning languages. It's been around for a long time. I think it's been around since the late '90s, and it's effectively a series of symbols that define planning.

What they were then doing was basically using PDDL to try and set a series of steps that the LLM would use to reason and get to an answer on doing a task or running an action. Then they tuned the PDDL using this tuning framework that they developed, giving it feedback, and they also fed it good plans and bad plans and said, “This is a good plan. This is a bad plan.”

Overall, the LLM was then run in such a way that it actually had a better set of steps that it would use to solve a particular problem. And so this can then lead to all of the underlying machinery of an LLM being better utilized to solve a bigger problem, to solve a chain of thought, or to solve some reasoning problem that requires several steps or planning.

I think it was a very good breakthrough. The benchmark data that they shared was pretty impressive, and it's getting quite a bit of attention this week. That was 1 really interesting paper that came out this week.

6. AI Moves To The Edge

The other one—and Nick, maybe you can pull this one up—this one's really impressive, Sacks. This comes from a team in Germany. This paper was published in the journal Nature Computational Science. These folks took a GPU, and for each token, typically you'll have the entire key-value cache transferred from high-bandwidth memory to cache memory, so this means that you're moving a lot of data between one type of memory and another type of memory. What they were able to do was reduce the physical memory size that's needed to run the attention window.

As a result, the energy and total token cost to run inference went down significantly. I'm trying to simplify this down as best I can, but what matters is the end data that they provided. Their architecture led to a speedup of 7,000× compared to the NVIDIA Jetson Nano, 300× compared with the NVIDIA RTX 4090, and 100× compared to the NVIDIA H100. The energy was reduced by 40,000× compared to the Jetson Nano, 90,000× compared to the RTX 4090, and 70,000× compared to an H100 for the same outcome.

So I think that this new kind of technique, if it scales, could have a pretty dramatic effect on the energy consumption needed to run AI. Importantly, because you need far less memory, you can move a lot of AI inference to the edge of the network. You could put, for example, a very high-powered LLM model that could be run in a robot, a piece of equipment, a computer, or on your phone—something that historically you'd need to run in a data center because you needed a very high-powered GPU chip stack.

This architecture, I think, could be one of these big architectural breakthroughs. We've spoken with Sergey Brin, Eric Schmidt, Sundar, and Demis about the big architectural breakthroughs that are coming in AI that could ultimately lead to many orders of magnitude reduction in the energy costs needed to run inference and AI models. Again, if this scales, then all of our assumptions about the data center and energy can start to be thought about under this new kind of architectural framework, which might naturally result in much, much lower need states.

We'll see, but it was a really important paper, and folks are going to look up this paper and say, “This could be a pivot point in how we think about the energy and infrastructure needs to support AI.” I don't know if you and your team have reviewed it, but it's definitely worth spending some time on.

Yeah, I think the writing was on the wall that models were going to get smaller and more efficient, to the point where they can run on the edge, on local devices. That was one of the implications of DeepSeek. But if you look more recently at the launch of Llama 4, their smallest model, I think it's called Scout—

Mm-hmm.

—runs on a single GPU.

Right.

So I think we're going to have a whole range of smart devices that will have a single GPU running a pretty decent AI model. Obviously, your phone will have one too, probably a much better one.

Have you and your team talked about what the energy demand curve looks like as these architectures get better? If we're talking about a 10,000× reduction in energy to run a token, have you guys thought about whether energy scales as we've projected it to scale? Do data centers need to scale like we've projected them to, or do you think that because they're more efficient, we'll actually have more demand?

That just sounds a little too good to be true right now.

Right.

Between papers and products, I pay a lot of attention to the launch of products. I don't pay a lot of attention to papers. I know that some papers end up being really important. For example, the paper on the Transformer architecture back in 2017 turned out to be enormously important. But I think that a lot of papers just don't really go anywhere for whatever reason. Maybe they're hard to reproduce, or they don't scale, or what have you. So I just don't really pay that much attention to the academic literature.

I do pay a lot of attention to product launches. When someone launches something revolutionary, it immediately gets everyone's attention because you don't have to speculate about whether a proof of concept is going to be possible. You actually see it. I guess what I'm saying is that the proof of the pudding is in the eating.

Right.

I think we're going to need a lot more power, a lot more electricity. I think that's pretty well known. We haven't even gotten to the robot revolution yet. That's coming in the next 5 years. That's going to be energy-intensive.

If this thing's even close to being correct, then you could run the most sophisticated LLMs in a robot without it needing to be run out of a data center going forward, and the robots could simply make a request for information from the internet when they need it. But all of—

Mm-hmm.

—the actual computation, the reasoning, all of the base knowledge would sit locally in that device. It's really incredible to think about. We are going to end up with these robots. It's amazing.

Yeah.

Yeah.

Well, I think that's right. I think self-driving wouldn't work if you had to run all the inference on the cloud. It's—

Right.

—run locally, by powerful AI chips. Then obviously it can connect when it needs to. But no, I would expect that robots are going to have a local AI model.

Yeah.

Okay, cool. Well, that's it. Those were the 2 papers I thought were really impressive this week.

7. The YouTube Restriction Explained

Hey, Friedberg, what exactly happened with YouTube? Do you have an update on what happened with our episodes from All-In Summit that appeared to be—

Yes.

—shadow-banned? Tell us what happened there.

Thank you to the folks at YouTube. They actually worked all weekend to help us figure out what happened, and there was nothing nefarious. There was no shadow-banning going on.

You guys may recall that a couple of months ago, we stopped bleeping out curse words in our episodes, and we muted them instead. When we muted them, the YouTube algorithm still thought that we were saying the curse word quietly, and it still showed up in the YouTube transcript. When you have a curse word in a video, YouTube marks it as restricted, so it's not age-appropriate, and that's why it was getting the restricted mark.

When we went back, the episodes that did get restricted all had a curse word in them, and we understand clearly what happened. Going forward, we're going to use the bleeping again instead of just muting. It was very benign, not nefarious. YouTube did a great job supporting us. We went back and fixed all the old episodes, so they're all out of restricted mode, and we started reposting all of our Summit videos again. The conspiracy corner is closed on that one.

Well, hold on. Do creators know about this? If you have F-bombs in your show, do you go into restricted mode?

You know, that's a great question. We were talking to the YouTube product team about this. There's no easy way for YouTube creators to see that a video has been tagged as restricted. They need to fix that. They're going to fix that, they told us, and I think we should all continue to hold them to that because it's important that creators know why.

One of the questions we had for them, which we thought was a theory, was whether, if people report your video, it automatically goes into restricted mode. The answer is no. Reporting triggers a review separately, but the restricted-mode algorithm is distinct. When it comes to restricted mode being triggered, you don't know that it happened. You don't get a notice. You're not aware of it.

Right.

They need to address that, obviously.

They need to have a dashboard that shows you any kind of restriction on your videos and a reason code for why.

That's right.

You know?

And specific timestamps, because their engineers were able to pull it up for us, look at the timestamps, and point us to them, and we could see what happened. That should be apparent. They should present that to the creators so they know why they got restricted.

I think part of the argument was, “Well, restricted mode in YouTube isn't a big deal.” It turns out it is a big deal. We saw it in our traffic. We had a big drop-off because a lot of network administrators—the people who run the Wi-Fi at Starbucks, on your public bus and subway, or in your office—have a network setting called Safe Mode.

Safe Mode was originally designed to block porn or other not-safe-for-work content at work, but it also causes restricted mode to be blocked on YouTube. If you're in one of those public networks and you're trying to access YouTube and the video is restricted, you lose that entire audience. It turns out, I think, that videos getting tagged as restricted mode is actually a bigger deal than folks realize, at least I think it is.

They should do a better job surfacing things, and creators should be able to go back in and correct any issues that might be causing a video to be restricted.

Yeah.

But I don't feel like the policy itself was bad. I think there was an algorithm problem where their software didn't pick up that we had muted bad words, and it was more apparent previously when we bleeped them. So we're going to go back to bleeping until they fix the algorithm.

And was there any weaponized reporting of content, or do we just not think that was a thing?

No.

Okay.

We know that not to be true. We checked that. I checked that at the high level, and the answer is no. I think we feel very good about that. There's no mechanism either that, if people do blast reporting or they try to, as we used to joke, brigade you, it doesn't actually trigger anything, so it's kind of—

Got it.

8. YouTube Admits Censorship Pressure

And speaking of YouTube, there was a really important report out this week. I think we kind of knew this, but YouTube acknowledged that during the Biden administration—roughly 2019 to 2022—that it censored something like 1 million videos at the behest of the Biden administration. I guess that would have started in 2021, and they admitted that they were pressured by the administration. Zuckerberg had said the same thing about Meta, and the Twitter Files informed us about the same thing, but now YouTube has finally acknowledged that, and there was a big release on that.

I'll tell you my view on this. I think the censorship that happened during that era is very important to have happened because it has brought it to light in a way that now there's a hypersensitivity to it not happening again, and I actually think that's very good. So the fact that it happened has now created a real sense that, going forward, the policy limits and the boundaries are now clearer than they ever were. It's not just about the Trump administration, which I think a lot of mainstream media tries to make it about, but it really is about the importance of free speech and censorship and who decides what's objective truth or not. Going back to all of the COVID-era discussions, not allowing people to have discussions clearly is a problem. And speaking of this topic, I don't know if you saw this, but there is this kind of hate speech bill that passed—

Yeah.

9. California Targets Online Hate Speech

It passed out of the Assembly and the Senate in California and is now on Gavin Newsom's desk to sign. It would basically fine social networks for allowing content to show up on their social networks that the state of California deems to be hate speech. Whatever language or terms the state of California calls hate speech—and you could see how this could become a very slippery slope very fast—they can now fine a social media company millions of dollars. In and of itself, that could actually propagate a whole new censorship regime where people who are using certain terms that in that era are considered bad terms or hate-speech terms are afraid that they don't want to get fined tens of millions of dollars, so they block all that content. And I do think that if this gets signed by Governor Newsom, it could trigger a whole new kind of censorship battle in the months and years ahead. We'll see.

Well, I think that's exactly right. I think the bill you're referring to is SB 771, and it is an EU-style suppression of “hate speech” on social networks. The problem is that there is no definition of hate speech. That's not a category that exists. It's just whatever the people in power say it is.

And so there is no constitutional exception for hate speech under the First Amendment.

That's right.

The California bill references civil-rights statutes, which speak to certain types of discrimination and certain types of hate speech. But the, to your point, those words are not defined. And so what ends up happening is you could say, well, using that word is discriminatory to this group in some way, or using that word is hateful because it offends another group. And suddenly you start to blur the line between what the average person might call hate speech and what perhaps some people—

Yeah.

—in an administrative body are calling hate speech. And suddenly it becomes more like, hey, is this really a civil-rights violation, or is it just offensive content? It's a very slippery slope that offensive content suddenly can get wrapped up and be called hate speech, and then the government starts to tell us all what we are and aren't allowed to see and say. We're obviously seeing the repercussions of that in the UK right now, where the police are knocking down doors to arrest people for putting stuff on Twitter.

The direction I thought you were going in a minute ago was that it sounded like you were saying that it's good that we've learned all these lessons from this COVID period where YouTube—

That's right.

—and Meta and X—

Yeah, then I changed my mind.

Yeah, and I think that's exactly right. I don't see any evidence that the—I'd say especially the political left—has learned its lesson. You've got Gavin Newsom now trying to ban hate speech in California. By the way, he also signed that bill—was it about a year ago?—banning parody. Remember that, parody videos?

I do remember that.

Yeah. There was an attempt to ban it—

It was political AI videos.

Yeah.

Yeah.

Because it was in the wake of a humorous fake advertisement for Kamala Harris. Then you've got these folks on the left already saying that Sinclair and Nexstar need to be punished for not putting Jimmy Kimmel back on the air. In other words, the same people who were saying a week ago that the Trump administration jawboned ABC and Disney, that that was fascism, but if they jawbone Nexstar and Sinclair, that's democracy. I mean, it's completely hypocritical. I'm not convinced anyone's learned a lesson from this.

And just to be clear, I don't think Jimmy Kimmel should be taken off the air or censored or whatever. I'm pretty sure that his show's not going to be back next year because it's got such low ratings. I don't think there's really a need to censor him. It is true that there is a public-interest requirement for using public spectrum, but nobody seems to agree anymore on what's in the public interest, so—

I completely agree with what you said last time. We just got to—

Yeah.

—auction off that spectrum. I don't think we need that spectrum. We have something called the internet now, and so we no longer need broadcast television. There shouldn't be a government-regulated broadcast television system where they're deciding what is and isn't appropriate content and in the public interest. That just doesn't make sense for the government to do in a market that's supposed to fully support free speech.

Right.

And I agree. I think the Jimmy Kimmel issue should be up to the people who are spending their money to put Jimmy Kimmel on the air, and they can decide what they want to do with him. If no one watches it, they'll take him off, and if people watch it, they'll keep him on. That's their decision. We shouldn't—

Yeah.

—I don't think it makes sense to, quote, cancel or ban someone for saying something that's offensive. And I think that if you do it on one side, eventually it'll happen on the other side. But that's a tried-and-true point in free-speech advocacy, obviously.

Yeah. I mean, look, you could definitely argue that throughout history, both the right and the left have sought to censor inconvenient speech when they've been in power. But again, I just think this is one of those issues where just because both sides have done it throughout history doesn't mean that in the present day, one side isn't a lot more guilty of it than the other. And you see this with whether it's Scott Wiener and Elizabeth Warren jawboning Sinclair and Nexstar to keep Jimmy Kimmel on the air, or whether it's Gavin Newsom possibly signing this bill to ban hate speech in California.

Well, do you think he'll sign it?

I don't know.

If he does not sign it—if he vetoes it—will you give him credit?

Yeah, for sure.

Yeah.

That'd be a good sign.

Yeah. Like when he vetoed that AI bill, I thought that was fantastic, and I give him a lot of credit for that.

Yeah.

I think he's—not a completely unreasonable person. There are points when he realizes things have crossed the line.

Yeah.

Which, by the way, is scary, because then I think about who's the next governor. These things pass out of the Senate and the Assembly, and he's the only thing standing in the way. It's scary.

Gavin Newsom isn't even the craziest person on the left. I think we can all agree on that.

Yeah.

I'm not a fan, but there are way crazier people. But let's see. I suspect he will sign it—

Mm-hmm.

—because I think the left very much likes this sort of thing. You see it in Europe. You saw it in the Twitter Files. You see it in these acknowledgments that YouTube has just made. You see it in what Zuckerberg told Rogan about the censorship that Meta was pressured to do. I think there is a clamoring on the part of the left to silence speech that they don't agree with, and they do call it hate speech. It's speech they hate.

And the right, in a fit of pique when they're angry about the assassination of one of their heroes, are they capable of saying that Jimmy Kimmel should be taken off the air? Yeah. But is that something that's been broadly acted upon on the part of the right? No, it's not. The left is the one that's engaged in massive amounts of speech suppression over the last few years. It's not a both-sides problem. So I hope Newsom will veto that bill, but let's see what happens.

I hope he vetoes it. I think it would be an incredible statement if he did. But if he doesn't, I could see why this creates effectively a free option for him, for the Democrats, for whoever's in charge of administering California's statute, giving themselves basically a free option on whether or not to enforce it and how to enforce it. It creates a mechanism, and I just don't like that mechanism existing, obviously.

But I could see how a system in power can find this to be a good mechanism for maintaining power and influence, or at least influence over speech.

I think Newsom’s going to have to sign it into law, because that’s what his base wants, and he intends on running for president in a few years. And it’s not just the California base; it’s the overall base of the Democratic Party. I don’t think he’s going to want to risk alienating them.

And even if he harbors some qualms, which I’m not sure he does, he probably realizes he’s got the Supreme Court to back him up in the sense that they’re probably going to find this unconstitutional. They’re going to throw it out, but he can still send a message to the left that he’s with them, that he wants to suppress conservative speech just like they do. So I suspect he’ll sign it.