Sean Ramis (3:02)

Hey, everyone, it is llama4day. Our goal is to build the world's leading AI open source it and make it universally accessible so that everyone in the world benefits. Really excited about, you know, all the unlocks that we're going to get from Llama three and then Llama four and then Llama five, and I think that's going to translate into better products. All right, welcome everybody to Llamacon.

Riley Griffin (3:22)

This is a system meant to rival OpenAI's ChatGPT or Anthropic's Claude. But when the rollout landed, it fell flat.

Sean Ramis (3:32)

Llama 4 just dropped and it's been an absolute disaster.

Riley Griffin (3:35)

Llama 4 is lost and it fell flat internally. According to our sources, it had just become clear that Meta was not leading the pack among these AI companies. Our listeners may think of Meta as the parent company of Facebook or Instagram or WhatsApp, you know, these popular apps that have billions of users. But Meta is increasingly seeing itself as an AI company too, and it wasn't leading there. And so Zuckerberg, seeing feeling that he was falling behind in this race, immediately sprung into action. From his homes in Tahoe and Silicon Valley, he started personally recruiting $200 million to work at Meta. That is what we understand. Mark Zuckerberg is currently offering to join his Superintelligence team.

Garrett Devink (4:23)

Meta has extended offers worth as much.

Riley Griffin (4:25)

As $300 million to more than 10 of OpenAI's top researchers. And he was basically quietly building this new secretive team that we now know is Meta's superintelligence labs.

Sean Ramis (4:42)

Okay, what is this super secretive team going to do? I thought Meta was like, all about the Metaverse and that's why they changed their name to Meta. I am proud to announce that starting today, our company is now Meta. It sounds like they should have changed the name to AI to AI.

Riley Griffin (5:02)

Meta. AI. It's a really good point. I think when I speak with folks on Wall street, the Metaverse is always the pain point, if you will. It is something that still bleeds cash. It's not a big revenue driver. It hasn't proved itself to be the dream that Mark Zuckerberg once painted. Back in the days where it was still called Facebook outright. You know, they were in the AI game. They founded this Facebook AI research lab, now called the Fundamental AI Research Lab. They actually brought in this guy named Yann LeCun.

Sean Ramis (5:39)

There is no question, absolutely no question, that at some point in the future we'll have AI systems that are as smart as humans in all the domains where humans are smart, but will be as smart as human, if not significantly smarter, in all domains where humans are smart.

Riley Griffin (5:55)

A big thinker, award winning thinker on some of these things. And so if you were watching that at the time, you might have thought Meta was really well positioned to be the AI leader. But when OpenAI came out with ChatGPT, it blew everyone out of the water, including AI researchers at Meta who I've spoken with. You know, they'd been building their own large language models, but the approach was still rather academic. It wasn't packaged in a consumer product in the way OpenAI so successfully executed on ChatGPT. And when I speak to folks who were at Meta at the time, they just, they believed already OpenAI had gotten ahead and was taking a more deft approach to consumer strategy. So I think that's when the race began. But it really kicked into gear this April with what felt to many internally at Meta as a flop with its latest release of llama.

Sean Ramis (6:57)

And what exactly has Meta done in this space so far? Nothing.

Riley Griffin (7:00)

It sounds like right now we're seeing a talent war. We're seeing Meta create a new organizational structure for its AI talent. So this is a multi billion dollar effort and Meta has the cash to deploy here. The perspective of company leadership is that they're looking to make a big bet and plug that talent gap. So. So Zuckerberg is pledging to spend not a couple billion, but hundreds of billions of dollars on their broader AI strategy and the infrastructure needed to support it. And yet they believe there's only a small talent pool that can best make use of that spend to put together these competitive models. Many are focused on reasoning, and that means they're specialists who can help build models that think step by step, as opposed to in this probabilistic approach, where they're predicting the next word in the sentence, because a lot of the Chatbots right now, they look like they understand what you're saying when you type in a question, but really they're just predicting. They've taken these massive amounts of data and they're able to predict the next word in the sentence. And what meta, what OpenAI want to do is really give them the capacity to think, to reason for themselves to get to that answer. So if you're going to spend hundreds of billions of dollars to the AI race, what's a couple billion on the talent that can steer the ship. But of course there's pushback. I will say that pushback isn't coming from Wall Street. I mean, Meta stock is up nearly 30% this year to date. Clearly there's excitement from the investor community at every given announcement on deploying cash to purchase companies to hire talent to build out data centers. More cash will be spent. One thing I've heard from folks who have left Meta over the years is that, you know, if you are a PhD type who got into this really complicated question of, of AI reasoning and superintelligence, you weren't necessarily inspired when Meta announced that they were creating AI chatbots. You know, that sounded like Snoop Dogg. That's a really different kind of product.

Sean Ramis (9:17)

You love puzzles and games.

Garrett Devink (9:18)

So do we.

Sean Ramis (9:20)

That's why we got Snoop Dogg as Dungeon Master.

Garrett Devink (9:23)

Your quest begins now, player.

Riley Griffin (9:25)

And so there has been skepticism of Mark Zuckerberg as a leader. He's someone who really wants to bring AI into the hands of individuals. And, you know, these folks who've departed Meta over the years point to the company as one that's flip flopped on its AI strategy. And I think it's too early to tell whether or not it's working. What has been successful is Mark's efforts to court some of the biggest names out there. So they are taking away top talent from competitors. But we also know there are people who have turned down Mark Zuckerberg's sports team level compensation packages.

Sean Ramis (10:07)

So it sounds like Meta is restructuring itself or creating altogether new structures to play the long game here on superintelligence. What does it look like in the rest of the industry right now?

Riley Griffin (10:17)

Everybody, I mean, everybody has had to respond to Mark Zuckerberg's crazed hiring push. His efforts have set a new bar for compensation across the industry. We've seen other companies like OpenAI have to make offers to their own employees to keep them that are much higher. You know, a rising tide lifts all ships. And there is no doubt that the way that they have aggressively hired and thrown cash at the problem has left other companies having to catch up. And you're even seeing that in the language, the public posturing from CEOs. OpenAI has come out and said that they're going to spend trillions on the AI race. You know, other companies are being quite loud about the money they're going to throw. But we're talking about the biggest players here, Sean. We're not talking about the upstarts that have just gotten some venture cash. Meta, Google, even OpenAI are in a different position because they have cash to burn.

Sean Ramis (11:30)

Read riley griffin@bloomberg.com Listen to us ask if Mark is wasting his money on Superintelligence When Today Explained returns.

Garrett Devink (11:50)

Million dollars isn't cool, you know it's Support for.

Sean Ramis (11:53)

Today Explained comes from Constant Contact. There are a few ways you could handle marketing your business. You could grab a giant microphone. I have a regular sized one in front of me right now. You go to a local mall, miss the mall and start screaming until mall security shows up. Or you could check out Constant Contact because they have better ideas than the mall. One Constant Contact says their marketing platform is here to make marketing easier and more effective for small businesses. The best part? They say you don't need to know anything about marketing. With their all in one platform, you can create and manage attention grabbing campaigns in just a few clicks. Email, text, social media, events, landing pages, you name it. It's all in one place. And with the help of their AI tool which they say helps you turn a rough idea into a ready to go message fast, there is no more juggling dozens of different apps. You also get automated sending, real time reporting and tools that actually help drive sales. You can get a free 30 day trial when you go to constant contact.com try constant contact free for 30 days at constant contact.com constant contact.com.

Garrett Devink (12:58)

Running a.

Riley Griffin (12:59)

Business comes with a lot of what ifs, but luckily there's a simple answer to Shopify. It's the commerce platform behind millions of businesses including Thrive Cosmetics and Momofuku, and.

Garrett Devink (13:10)

It'Ll help you with everything you need.

Riley Griffin (13:12)

From website design and marketing to boosting sales and expanding operations. Shopify can get the job done and make your dream a reality. Turn those what ifs into Sign up for your $1 per month trial@shopify.com SpecialOffer Labor Day Savings are happening right now.

Garrett Devink (13:30)

At the Home Depot. So what are you working on? Prep for fall with our wide selection of cordless power tools that make it easy to clear your lawn starting at $79. And once the leaves are clear, keep your yard looking fresh with colorful mums that bloom all season long. Shop Labor Day Savings now through September 3rd only at the Home Depot. See select stores for details. Cool.

Sean Ramis (14:01)

Today Explained Garrett Devink writes about tech for the Washington Post, and we asked him here to help us wrap our heads around this one big beautiful word, superintelligence.

Garrett Devink (14:13)

You sort of see it pop up here and there. I mean there's different terms. Ultra intelligence, superhuman intelligence. I think the actual word superintelligence was used by Nick Bostrom. Cures for aging, space colonization, self replicating nanobots, or uploading of minds into computers, all kinds of science fiction stuff that's nevertheless consistent with the laws of physics. All of this superintelligence could develop, and possibly quite rapidly. Who was this technologist and sort of futurist who spent a lot of time writing and talking and inspiring people about the singularity, the future, when brains meld with machines and we become infinite beings. Once there is super intelligence, the fate.

Sean Ramis (15:00)

Of humanity may depend on what the superintelligence does.

Garrett Devink (15:03)

Machine intelligence is the last invention that humanity will ever need to make. A lot of people who have been working on AI over the last decade have made their goal to invent a superintelligence, something that is smarter than humans. And it does bleed into marketing. It is about getting people along, but it's also something that a lot of people actually believe, and they think it's coming.

Sean Ramis (15:28)

What all the AI companies are aiming to build is digital superintelligence. So an intelligence that's far smarter than any human, and then ultimately an intelligence that is far smarter than all humans combined.

Garrett Devink (15:41)

Some people argue, oh, well, we already have artificial general intelligence. That's what ChatGPT is. Look at all the cool things it can do. Other people say, don't be ridiculous. It's making all these mistakes. It's just a random number generator that we've trained to do cool tricks. That's kind of where the AGI debate is. And so I think people have now glommed onto the term superintelligence because it is something that we can all kind of agree we're not there yet. We don't have some kind of oracle in a box that is inventing new forms of science that we don't understand yet. And so we can kind of say, that's that's what's coming, and give people an idea and a vision to work.

Sean Ramis (16:22)

Towards, even if it could be decades away.

Garrett Devink (16:27)

Yeah, and I think here in Silicon Valley, there's very few people who are out there saying it's decades away. Right. I mean, that's where they were a few years ago, before some of the breakthroughs that trigger ChatGPT and some of these other AI products that we're all dealing with and talking about right now. But most people here in Silicon Valley, in the tech industry, have rapidly moved up their timelines for when they think we will invent a machine that can do things that humans don't understand, that can go beyond us.

Sean Ramis (16:58)

Developing superintelligence is now in sight At Meta, we believe in putting the power of superintelligence in people's hands to direct it towards what they value in their own lives.

Garrett Devink (17:07)

So you have people saying 2026-2027-2028-2030. And they do kind of, you know, there were a few 2025s. We're in 2025 now, and the 2025 people are saying 2026 now. So I do think there is a bit of an effect of, you know, there's definitely a lot of optimists here, but that's sort of definitely a big change in how people who work on AI think about the capabilities and how quickly they might be approaching.

Sean Ramis (17:37)

We recently did like a two part series on Mars on this program. And it's funny because that is reminding me of Mars where Elon Musk was saying, like, probably got about a 50% chance of sending ships from Earth to Mars at the end of next year. So November, December, next year, in about 18 months.

Garrett Devink (17:52)

Wow.

Sean Ramis (17:53)

And it's just absolute pie in the highest of skies. Does anyone who's promising superintelligence in the next few years, or even this year, it's August, by the way, do any of them have any idea what this thing's gonna look like?

Garrett Devink (18:10)

You're starting to see people get the questions exactly like that. Like, what are you talking about? This is science fiction. Like, you have no evidence that this is coming, that it's happening. Just because something is advancing relatively quickly or faster than you had expected, doesn't mean that that's gonna continue. Right. They're sort of extrapolating into the future without knowing for sure that that's gonna happen, or even presenting a strong explanation for how it might happen. So those people in the tech industry, I mean, we're talking about, you know, leaders at OpenAI, at anthropic and Google, also people who spend a lot of time researching AI and how it might affect society and the economy have obviously been getting these questions. And so they're starting to refine their answers a little bit. And I think one that is quite interesting and has taken hold is the CEO of Anthropic AI. This is a, you know, very, very prominent, well funded AI company here in San Francisco. Dario Amadai, the CEO, he started talking about a nation of geniuses.

Sean Ramis (19:10)

In terms of pure intelligence. It is smarter than a Nobel Prize winner. Across most relevant fields. Biology, programming, math, engineering, writing, etc. Each of these million copies can act independently on unrelated tasks, or if needed, can all work together in the same way. Humans would collaborate. We could summarize this as a country.

Garrett Devink (19:35)

Of geniuses in a data center, and they are going to invent these scientific breakthroughs, or we'll give them complex engineering problems, or even societal problems. We'll say, hey, how do we ensure that democracy stays strong into the 21st century? Those are the kinds of more concrete, yet still very vague ideas that people are starting to articulate.

Sean Ramis (20:03)

Is there anyone saying, this is a crock, it's not going to happen? Is there anyone saying, you guys just made this thing up because it sounds good and it raises money?

Garrett Devink (20:13)

Absolutely. And I think there's a very strong argument that that's all it is. I mean, if you look at the last decade, 15 years of silicon Valley, obviously you think about meta Facebook, and the same people who built these companies are now building these AI products and trying to find ways to convince people that we need them, that we need to use them, that we need to work them into our daily lives, and yes, that we need to spend money every month in order to continue to access them. And so the other element too is if you have people talking about science fiction and AIs coming to destroy humanity, you don't have them talking about racial bias being baked into AI. You don't have them talking about how AI is being used to make it easier for the government to surveil citizens. And so I think there is a strong argument that a lot of this is. Is about marketing, is about hype, is about distraction, really. And there are definitely people who are articulating that argument who are saying there's a big bubble. These companies are way too overvalued. They're not actually delivering on what they're claiming. They're all going to, you know, blow up, and the stock market is going to go with it. And I do think that there's. The truth probably lies somewhere in between. I mean, living here in San Francisco, it might be hard to sort of really understand from the outside, but there are people who truly, truly believe in their heart of hearts, almost in a religious way, that superintelligence is coming and that it's our job as humanity to prepare ourselves for it.

Sean Ramis (21:44)

And in the meantime, we peons get like, GPT5.

Garrett Devink (21:47)

Yeah. I think GPT5 is a really interesting story that reveals a lot about where the technology industry is right now. And so GPT5 is the latest version of OpenAI's technology. It's sort of the software that goes behind ChatGPT. And a lot of people were really expecting that GPT5 would blow everyone away. That it would solve a lot of the problems that AI has. You might have heard about hallucinations. That's when AI just makes something up randomly and acts really confident that it's true. When it isn't, people said, oh, GPT5 hallucinations won't really be a problem anymore. And when it actually came out, it was better at most of those things, but it wasn't a huge leap in the way that people were expecting. And I think that huge hype cycle, all that energy and all those claims by companies like OpenAI, really fed this expectation amongst people who are interested in AI, are early adopters, that it was going to be the next big thing and all the skeptics and naysayers would be proven wrong. And when it came out, it's not like it was worse, but it definitely wasn't super intelligence. And so you see a lot of soul searching going on after GPT5's launch.

Sean Ramis (23:01)

If you're just, you know, a person who occasionally uses ChatGPT to like, plan a trip or, I don't know, write a speech at a wedding, which I've heard over and over that people are doing. You know, like, how. How should a regular person who uses tech casually to, like, help their life be a little more efficient be thinking about superintelligence, or should they not be thinking about superintelligence at all?

Garrett Devink (23:28)

I think the first way that people should think about it is with skepticism. They should see it as a marketing term. They should understand what these tech companies are doing. And I think beyond that, if, you know, if you care about superintelligence, you care about AI. I mean, it's an interesting thing to think about. There's a reason people are drawn to this, because science fiction, we have all these ideas from literature, from film that are very exciting and interesting. And I do think it's fun to kind of engage with these questions. It's kind of cool to live in the future, I think, especially as a tech reporter. We were writing about how bad technology was and social media was destroying society. We sort of spent a lot of years and that was the big story. And now the big story is, yes, there's a cynical, skeptical approach, which is tech companies are trying to sell us something that we might not even need. But at the same time, the technology is advancing, it can definitely do things that we couldn't do just a couple years ago. And we sort of see the future unfolding in front of us a little bit. And so I don't think there's any harm in being excited about that and looking forward to ways that technology might make our lives better, as long as people are aware of some of the motivations and incentives behind, you know, the things that the companies are saying.

Sean Ramis (24:57)

That was Garret Devink from washington post.com he spells it G E R R I T. How do you spell it? Our episode was mixed by Andrea Christensdottir and Matthew Billy, edited by Miranda Kennedy and fact checked by Today, explained senior researcher Laura Bullard. It was produced by Gabrielle Burbe, who was off to pursue her Bay Area dream not to work for Metamark, but to be a bonafide reporter. Happy trails, Gabby. Those who remain are Patrick Boyd, Rebecca Ibarra, Abishai, Artsy, Miles, Bryan, Peter Balin On Rosen, Devin Schwartz, Hadi Mwagdi and Denise Guerra, who helped out with today's show. Danielle Hewitt joined the team this week. Welcome, Danielle. Noel King has been here for years. Aman Al Saadi has been here even longer. And Jolie Myers isn't thinking about the news this week. Or maybe she is. It's hard to say. We use music by Breakmaster Cylinder. I'm Sean Ramis from Today Explained is distributed by wnyc. The show is a part of the Vox Media Podcast network. You can find many more of our shows@podcasts.voxmedia.com.

Garrett Devink (26:09)

Sam.