Riding Llamas and Wakeboards
Kurt Wagner
As noted in yesterday's post about Mark Zuckerberg's "Thoughts on Open Source AI", I thought it was separately worth linking to the above post – and really, the entire video below – where Emily Chang sits down (well, and wakeboards) with Zuckerberg. Chang does a nice job asking him about topics he either breezes over or doesn't wade into in his own post on AI. Notably, OpenAI, the startup that pretty much everyone would consider to be the current "leader" in the space.
As Kurt Wagner relays from the chat:
Meta uses its Llama models to power its AI chatbot, called Meta AI, which operates inside its apps, including Instagram and WhatsApp, and also as a separate web product. Zuckerberg said that Meta has “hundreds of millions” of users for its chatbot, and expects it will be the most widely used chatbot in the world by the end of the year. He expects that others outside of Meta will use Llama to train their own AI models.
Highlighting is my own, of course. But that's the key bit here as it pertains to OpenAI. People seemingly get confused by Meta's AI approach and strategy because there's Llama, which was the focus in the post and announcements yesterday, but there's also "Meta AI" which is Meta's own AI product (read for now: chatbot) built on top of Llama. And that's the product which Zuckerberg is predicting will surpass all other chatbots – notably, of course, OpenAI's ChatGPT – by the end of the year. And actually sooner, he predicts to Chang.
I wrote about Meta using their scale advantage back in April, and clearly it's working at least from a raw numbers perspective, with Zuckerberg saying there are already hundreds of millions of users of Meta AI. That's what happens when you shove it into not just one, but multiple services with a billion-plus users. To be fair, it's not live everywhere around the world yet – including here in the UK, and perhaps it never will be in the EU! – so they haven't fully pushed that advantage yet. And the actual results for how people are using such tools thus far are mixed at best. Still, make no mistake, this is Meta going right after OpenAI, just as they would and have done to any other competitor.
When Chang asked Zuckerberg directly for his thoughts on Sam Altman, he said he deserves a lot of credit for what he's done with OpenAI to date – especially with all the noise and drama surrounding both himself and the company, something which Zuckerberg jokes that he knows about. But then he takes his shot, noting that it's ironic how a company named OpenAI is building closed models. Everyone already saw Zuckerberg attempt to get some brownie points with "badass" Donald Trump (which happened a bit later in this interview), but this swipe seems like it may have scored some points with Elon Musk too. (And it has.)
But also worth calling out is the very last line in that excerpt above. Zuckerberg mentioned this in passing in his post, but the notion that a key element of Llama will be for third-parties to use it to train their own models. He believes that custom models will be the key for different workloads and that Llama can help underpin them all – millions or even billions of them, he says at one point.
This goes back to his notion stated previously that there will not be "one AI to rule them all". Which, yes, is talking his own book, but sounds reasonable enough. Though I'm still not sure how you square the idea that he's also chasing after "AGI", which Chang pushes him on and he admits that there's not a single definition for what that is. But he thinks Llama 4, which he keeps mentioning is in the works – presumably because OpenAI's GPT-5 is also in the works and likely to be unveiled at some point soon – will get us closer to it. Whatever it is.
Zuckerberg said that Meta’s Llama 3 models cost “hundreds of millions of dollars” in computing power to train, but that he expects future models will cost even more. “Going forward it’s going to be billions and many billions of dollars of compute” power, he said. Meta in 2023 tried to reign in some of its spending on futuristic technologies and management layers, cutting thousands of jobs in what Zuckerberg dubbed the “year of efficiency.” But Zuckerberg is still willing to spend on the AI arms race.
“I think that there’s a meaningful chance that a lot of the companies are over-building now, and that you’ll look back and you’re like, ‘oh, we maybe all spent some number of billions of dollars more than we had to,’” Zuckerberg said. “On the flip side, I actually think all the companies that are investing are making a rational decision, because the downside of being behind is that you’re out of position for like the most important technology for the next 10 to 15 years.”
Also a reasonable stance, though I'm not sure how much longer Wall Street is going to agree – and actually, they already don't seem to with Meta itself. Not that such things have stopped Zuckerberg before (and good for him – a stance he can take thanks to his control of Meta through voting shares)!
He also takes his swipe at Apple in his chat with Chang, though not quite as directly as he does in his post. Though he does manage to get a second shot in when asked about the mental health crisis – he pushes back on the notion that social media has much to do with that and suggests that perhaps smartphones themselves are more problematic.
Lastly, when asked to revisit the renaming of the company to Meta, Zuckerberg admits that he thought AR would be here before AI, which almost implies that he may have picked a different name. Can you imagine the artist-formerly-known-as-Facebook being called, say, 'Intelligence'?1 It sounds ridiculous, but so did 'Meta'! He also notes it was an obvious way to ensure they didn't get pigeon-holed as just a social media company. Which does seem to have worked.
The whole chat is well worth the 24-minute watch.
I can't get enough of these 'Zuckerberg riding a Llama' AI images. If only NFTs were still a thing...
1 Being called 'Artificial' is unlikely something the company would have chosen, though many would deem it to be more appropriate.