The Fight Tearing Apart Silicon Valley

featuring the voices of our writers and columnists.

To me, the single most underestimated force in international relations is actually stupidity. Including Jamal Bowie, Trusty McMillan Codham, Newfelt Goldberg, Thomas Friedman, and many more. Find the opinions wherever you get your podcasts. From the New York Times, I'm Natalie Kitrowev.

This is the Daily. For years, a debate has been simmering and Silicon Valley over the right way to build AI. Should the technology be open and free for anyone in the world to use,

“or should we tightly control how it's made and who can access it?”

This week, that debate reached a boiling point, dividing the country's top AI companies into two opposing camps, and drawing in some of the biggest names in tech. Today, my colleague Mike Isaac, on what started that firestorm, and how it could end up reshaping the future of artificial intelligence.

It's Friday, July 31st. Mike, welcome back to the Daily.

It is always a pleasure to have you here.

Hi, thanks for having me back. So, you actually played a not insignificant role in this AI debate, blowing up bursting into the open this week. So, just start by telling me what happened. Yeah, so on Tuesday morning, I'm doing my normal coffee run

to go to Starbucks up to Hill for my house, and got my Vinti, I used to Americano, and my hand, well, two of them, actually, because I, too. I know it's really embarrassing. I have one saved for later. Wow.

So, I get a ping from someone at Meta saying, Mark is going to call you pretty soon, and I realize they're talking about Mark Zuckerberg. And we should just say, right, it's a pretty big deal that Mark is calling you, Mike.

I mean, this is not a guy who typically talks to news papers these days. Yeah, I mean, he's not a huge fan of the mainstream media these days. I have not talked to Mark Zuckerberg on the record and at length on a phone call for three years, I want to say, you know, the point being,

when I get a call from Mark Zuckerberg,

“I know it is important, and something he really cares about.”

And we start talking, as I finally get in front of my computer,

and it is immediately clear to me, he's there to start chucking grenades at his competitors. What do you mean about that? So, the polite way of doing it is to not mention anyone by name, but it is very clear he was aiming squarely at Enthropic and Open AI,

which are two of the biggest technology companies in Silicon Valley, and widely considered the leaders in artificial intelligence right now. And so, his point of view was these two companies have too much power and too much say over where this is headed

in terms of how we build AI. And more people should have a voice in building that next frontier of technology.

“And I think while it is a sort of self-interested argument”

he's making, it also aligns with the way a lot of other people in the industry thing. Okay, and this folds into a debate that has been very public over the last few weeks, just give that context. Layout exactly what he's actually talking about. Yeah, so, in recent weeks,

we've seen a bunch of Silicon Valley companies really split into two camps over how AI should be shared and developed. In one camp, you have Open AI and Anthropic, which are both saying AI is super dangerous, and we need to keep a really tight lid on how we develop it,

and who can access it just to be safe. And then you have, in the other camp, companies like Meta, Microsoft, and Video, who are saying, "Actually, we need to share this technology really widely, and clamping down on it only serves to benefit the companies that are in the lead

and the AI race, namely Open AI and Anthropic." Now, I should say this debate over the sharing of tech, and whether it should be open or closed is a longstanding one across Silicon Valley.

And Mike, why exactly did this argument erupt right now at this moment? What actually kicked it off? I think this debate really kicked into overdrive a few weeks ago

with the introduction of this powerful new Chinese AI model called Kimmy.

And basically, whenever a new AI model comes out, what researchers do is put all the data out there as far as the benchmarks, how it's performing, kind of to prove, "Hey, we did something really good and check out how cool it is." And so the reason it blows everyone away is because it basically performs almost on par in certain areas

with the leading models out of Open AI and Anthropic. And that's a big deal because this model is based on open source software. And what exactly is open source software for those of us who may not be as well versus you?

“How much time do you have to get into this?”

About 30 to 40 minutes. Okay, I'll make it the sharp version. Open source software is about as old as Silicon Valley. And it's this idea that code that runs programs should be given away freely and shared between people

so that they can build on it or use that to create new programs. It shouldn't cost anything. You get to see everything on how the programs run so that you can poke holes in it in case it's insecure. It's like the coding equivalent of showing your work.

Essentially you put this stuff out there and people can tweak it. They can modify it. They can build their own stuff with it. Yeah, 100%. It's a fundamental tenant.

“I think of a lot of technologists out here because they really believe in the free sharing of information”

and building new exciting programs using this code. And look, in China, this is basically almost a national policy to embrace open source software. Yeah, and that gets at one of the questions that I have, Mike, which is why would a Chinese AI company or any AI company for that matter just give this away for free?

Yeah, I mean, it sounds counterintuitive, but there is a business strategy to it. If you give your tech away for free, I think you have a better chance of people trying it out and ultimately adopting it. You start gaining market share when you might not have had any users before.

And there are ways to make money on that later. It's kind of an old idea in the tech industry of you get the users first

and then you can worry about the money later because it always comes.

But it also gives you the power to shape where the tech is going because everyone is now building on your stuff. Right. And that's essentially what the Chinese companies are doing. They're flooding the zone, getting more people to use their tech

instead of paying open AI or anthropic to use their models. Right. And for those leading companies in the US and anthropic and open AI to have a competitor that is making what you're making just about as well as you are and handing it out for no money so that it can spread more widely. I mean, that's a pretty big threat.

That is an existential threat, I would say. And so how do these companies open AI and anthropic react to Kimmy? I assume not well? Yeah, that's an understanding. I think they have already seen some of their customers

turning to these open source alternatives and that's a huge threat to their business. Right. But I also think there's a sort of higher level philosophical concern around open source software and specifically around security, which is explaining that to me.

Look, open source software, basically anyone can download it and use it. There's not really any guardrails on that. What folks at anthropic and open AI are a bit worried about is, you know, what if the Chinese government wants to use that for ill against the United States?

What if some random hacker anywhere can download a very powerful model and start spreading disinformation or doing mass scams or use it to hack

“any number of very important companies that run the infrastructure globally?”

What if they cause havoc on Wall Street by doing some crazy auto trades or tracking a bank?

They're basically saying, when you make this code available for such powerful tools

in AI, you run the risk of essentially handing nefarious actors the keys to what could be really dangerous weapons in their hands.

That's how they talk about it often.

“I think some folks see the more cynical side of it and say,”

it's really a business model threat, I do actually think there are a lot of true believers who say this stuff is dangerous and it shouldn't be handed out willy-nilly. And so what are these companies and these industry people who are worried about this do about that fear? Folks at Open AI and Anthropic have been talking to regulators in Washington for a long time.

You know, they've been saying, how do we develop a framework to make AI safe? They dispatch, you know, their top lobbyists are talking to folks inside of the White House inside of different sectors of the Trump administration that have been working on this for a long time. And these talks have been going on for a while, and they essentially want to pick up the pace and say, let's talk about what actual restrictions

around Chinese models could look like. And so to make their case, they actually point to a crazy incident that happened inside of Open AI that week, which is. So all the time, I meant opening either testing these advanced models that are unreleased and, you know, they say are really crazy, you can do all this stuff. So they have this unreleased model that they've been testing inside of the company.

And they're trying to essentially push it to see if it can carry out some hack. Usually what they do for these scenarios is put it inside a box so that it can't get out and wreak havoc across the world. But despite all of their controls, it ends up breaking out of the box anyway. Wow, their model decides to find a way to access the internet. Go find this other company.

It's called hugging face hack into the company and dig around for a few days until both Open AI and hugging face. Figure out that this model broke out and started hacking. It's kind of the nightmare scenario that a lot of researchers talk about all the time. Like, oh, man, our stuff can go crazy and do what it wants. Right.

It's like a scary sci-fi robot taking things into its own hands. But that's an Open AI model, not a free Chinese one.

“So how exactly does this help them make their case against these open source competitors?”

Well, it's funny. Like, obviously, Open AI doesn't want their stuff to go hack other companies. Right. There's questions of whether it's like a felony or not and do charge a computer with a crime. But it provides a data point for them to go to regulators and say, hey, we were right.

This is actually really dangerous. If these models that we're developing in a closed environment can do this, just imagine what open source models backed by another government in China that is not necessarily aligned with the American interest could do to the world. You need to listen to us right now.

They also positioning themselves as the only essentially responsible stewards of developing this technology. They should be the ones in charge of it.

“And how does that argument go over in the industry?”

I mean, I have not seen an explosion online like this in quite some time in Silicon Valley.

Basically, everyone else from CEOs of the biggest companies to all of venture capital flip out on these two companies.

And they make, what is essentially the exact opposite argument? It was basically opening eye and anthropic against everyone else. We'll be right back. I'm Dan Barry and I'm a longtime reporter with The New York Times. I've been here for 30 years and I've seen a lot of things change.

I was here before there was a website. But one thing hasn't changed at all, and that's the mission of The New York Times, to follow the facts wherever they lead. And if that means publishing something a government or a leader or a celebrity doesn't want to aired, that's not our concern.

I've never been told to go against the facts to accommodate anyone.

And if I had, I would have quite frankly left the building. This is the way it was when I was covering the aftermath of 9/11. And this is the way it is now as I cover the United States of today. If you believe in the importance of factory and reporting, you can support it by becoming a New York Times subscriber. And if you already subscribe, this veteran reporter, thanks you.

Okay, so this is a moment of high drama in Silicon Valley where you are right now. Bring me into it. Like, what exactly does this freak out look like? And what is the case that all of these CEOs are making against open AI and anthropic?

Just for context, if you don't know who Jensen Wong is and that would get you some side eyes and Silicon Valley, if you don't know Jensen Wong is. He is kind of the king of the valley right now. He's a CEO and founder of Nvidia.

They make the crucial AI chips that run these programs.

They sell billions of dollars worth of those chips to basically every AI company on the planet. One of the most valuable companies in the world. Yep, right now and everyone loves him like he's got he also wears like a leather jacket literally all the time. It's very funny. I like very easy way to get people to love you.

“I don't know if I could pull it off, but yeah, that's what he does.”

So look, he did not have an X account until last week when he decided to post his first ever social media post on X. To raise his voice on the issue essentially. And what does he say? Let me read for you.

For my first post, I'm sharing a letter in video signed on why open models matter.

AI will transform every industry power every company and be built by every country. Open models strengthen safety and cybersecurity, accelerate innovation and diffusion and enable sovereignty. The world needs both frontier closed models and frontier open models. Okay, so this sounds like a pretty full-throated argument for open source models, right? Yeah, that's right.

And I think that is really what broke the damn wide open. After we see Jensen's tweet circulate. By the way, this was like it like 6/3 in the morning, my time. Everyone in Silicon Valley, Sachin Adela, the CEO of Microsoft, he tweets a similar message nine minutes after that. Elon Musk replies, "Jensen is right. This has my full support."

Huge voices in venture capital, Bill Gurley, this legendary VC, Mark Zuckerberg later in the day, does a quote of Jensen's post and is like, "Yep, I'm on board." Basically, everyone in the valley bans together and says, "Rara, open source."

Frankly, I've never seen anything like it in quite some time.

“Did you have the sense that any of this was coordinated beforehand or was it spontaneous?”

Oh, no, for sure. There was definitely a lot of like back room. Okay, we need to get together and stump for open source. I mean, I could be wrong, but I don't think Sachin Adela has fired up tweets at 6/30 in the morning, but who knows, like, maybe they're really risers. And how did these industry titans respond, if at all, to the argument about safety,

to the concern that the bad guys could use this open source AI software for nefarious purposes? I think this goes back to some of the core arguments for open source to begin with, which is the idea that if you develop software out in the open, it can make things safer for everyone. You can pick it apart, more people's eyeballs are on it.

It's kind of a egalitarian view of what software development should look like.

“But I think there's also an economic incentive for these guys to make that sort of argument as well.”

Right, just explain that. It is not good for the rest of the AI industry. If two companies are essentially dominating the whole thing in the United States. In video, for instance, they sell chips to everyone. The more people using different types of AI programs,

whether they're open source or closed, the better it is for companies like in video. And similar for companies like Microsoft, which has huge cloud computing businesses that run all of these different types of open source models. They have a real economic interest in keeping this open. And then there's, you know, the rest of Silicon Valley.

There's the venture capitalists who did not invest in open AI and anthropic and one other companies to do well in the AI age. Everyone in the valley who's not open AI and anthropic are just worried that some sort of broad regulatory move will make them kingmakers for the foreseeable future. Okay, so these companies come out.

They're making a philosophical case for open source. But in the background, there is a clear business benefit that they all have in taking this up. They have a stake in open source.

to these tools to these models.

That would help everyone else catch up here. Yeah, I mean, this is like business 101. If there's two leaders in the lead and you have other advantages coming in in terms of cheaper open source models or broad availability of alternatives, that levels the playing field in a way that hasn't been level for a while.

And what about meta? Did you ask Zuckerberg about all this on the call?

“Meta in particular, I think, is like a really interesting kind of case study.”

A few years ago, when they were very behind in the AI race, Zuckerberg decided they were going to do open source

AI software development. They tried to catch up by doing exactly kind of what these Chinese companies are doing and spreading their software across the valley and hopefully the world. That stalled out. That plan wasn't working.

So they pivoted hard and spent billions of dollars essentially doing what open AI and anthropic are doing and developing closed models. And now it's funny because we're back in this spot where open models are the new hotness again and he's pushing those. So I tried pushing him on this point a little bit and saying,

“what is your strategy and can you at least acknowledge that you've done both of these things?”

And he's not quite willing to go there yet whether they want to go fully open or closed. But I think it was clear in this call that he was making a full court press. And it wasn't just on this call. That morning he pinned an op-ed in the Wall Street Journal. He called up the financial times later that day.

Mark, who very rarely does these mainstream sort of pushes for what he cares about, at least in the media, suddenly is going all out on this particular point. Can I just ask why exactly you think he was making this full court PR press? I mean, who specifically is he trying to reach? Who's his intended audience?

Yeah, I mean, look.

“I think all of these people really have kind of an audience of one or one group of people.”

And that's all of the decision makers in Washington. Right now, lobbyists are descending on Capitol Hill. The CEOs are going to Capitol Hill and trying to convince the Trump administration to lean in their direction. And kind of making the case either that this stuff needs to be way more secure or that open source is the way to go. And the latest that I've heard is the folks in Treasury and the White House and Commerce are curing what both of these sides have to say,

and are still in the middle of deliberating how they're going to come down on this whole thing. And those deliberations are happening in the context of this big competition between the US and China, right? The US does not want to lose the AI race to China and each side of this debate thinks that their way puts America in the best position to win that race.

The question seems to be which premise does Washington buy, ultimately?

I think Washington is in a really interesting and tenuous situation where they don't want to upset folks on both sides, essentially. My guess is that they're kind of trying to thread the needle for some sort of compromise, maybe not fully clamping down on open source software because that would tick off basically the entire industry. But perhaps some way of reigning in the widespread use of Chinese models, at least from American companies, but it's still kind of TBD on whether actually going to land.

But the stakes for lawmakers, for the people who are making these decisions, they seem pretty high. Like what they decide about how to regulate AI doesn't just affect these juggernaut companies, it affects the whole economy, which is so dependent on AI right now. And it affects a lot of us on a really personal level if we interact with these models, which is becoming increasingly unavoidable. I mean, that's exactly why everyone's being so loud about this right now. The administration has to think about security online. They have to think about Chinese models and whether those threaten national security.

And I think they have to think about protecting this kind of golden goose that is really the engine of the US economy right now. You know, you have construction jobs going up because people are building data centers. You have this talk market essentially growing because of all the gains and investments in AI. And so the wrong decision essentially could have huge consequences for the future.

Well, Mike, thanks for being here. We really appreciate it. Thank you for having me.

On Thursday and Thropic announced that it too had a security breach. The incidents that were eerily similar to what open AI reported just nine days earlier, the company said that during a routine test several of its AI models hacked into three outside organizations. We'll be right back.

“Here's what else you need to know today. President Trump said on Thursday that he may withdraw his nomination of his former personal lawyer Todd Blanch as attorney general.”

The nomination had stalled after Republicans postponed a key vote on the matter this week.

Republican senators John Kornin and Tom Tillis had threatened to derail that vote over a deal that Blanch had cut to shield Trump and his family from tax investigations. Kornin and Tillis are both not returning to Congress next year after Trump refused to endorse them in their Republican primaries.

“Trump said he could renominate Blanch after the two senators leave office next year, though it's not clear if by that point Republicans will still have their commanding majority.”

And the U.S. economy slowed its growth in the second quarter, weighed down by the war in Iran.

Economic data released on Thursday showed that growth over the past three months was 1.5 percent lower than in the first quarter of this year.

The report did show that overall consumption was relatively healthy and that business investment particularly tied to AI remains strong.

“Today's episode was produced by Diana Win, Chris Bender of and Stella Tam, with help from Adrian Hurst.”

It was edited by Annie Minoff and Mark George and contains music by Mariel Azano, Dan Powell, Rowan Nemisto and Pat McCusker. Our theme music is by Wonderley. This episode was engineered by Alyssa Moxley, special thanks to Kade Mets. That's it for the Daily. I'm Natalie Kitro left. See you on Sunday. Coming up on the Wirecutter Show. We're talking about forever chemicals.

We were really interested in looking into areas where PFAS have historically been starring in how the products perform. Cookware, cosmetics and waterproof gear. What they are, how we're exposed, and what you do and don't need to worry about in consumer products. Find the Wirecutter Show wherever you listen to podcasts.