
33 segments available
Zuck on: * Llama 4, benchmark gaming, open vs source * Intelligence explosion, business models for AGI * DeepSeek/China, export controls, & Trump * Orion glasses, AI relationships, and not getting reward-hacking by our tech 𝐄𝐏𝐈𝐒𝐎𝐃𝐄 𝐋𝐈𝐍𝐊𝐒 * Transcript: https://www.dwarkesh.com/p/mark-zuckerberg-2 * Apple Podcasts: https://podcasts.apple.com/us/podcast/dwarkesh-podcast/id1516093381?i=1000705423020 * Spotify: https://open.spotify.com/episode/7Brv4vg9P8a8CNvAN5UsMv?si=d548a8aef39e4b65 𝐒𝐏𝐎𝐍𝐒𝐎𝐑𝐒 * Scale is building the infrastructure for safer, smarter AI. Scale’s Data Foundry gives major AI labs access to high-quality data to fuel post-training, while their public leaderboards help assess model capabilities. They also just released Scale Evaluation, a new tool that diagnoses model limitations. If you’re an AI researcher or engineer, learn how Scale can help you push the frontier at https://scale.com/dwarkesh. * WorkOS Radar protects your product against bots, fraud, and abuse. Radar uses 80+ signals to identify and block common threats and harmful behavior. Join companies like Cursor, Perplexity, and OpenAI that have eliminated costly free-tier abuse by visiting https://workos.com/radar. * Lambda is THE cloud for AI developers, with over 50,000 NVIDIA GPUs ready to go for startups, enterprises, and hyperscalers. By focusing exclusively on AI, Lambda provides cost-effective compute supported by true experts, including a serverless API serving top open-source models like Llama 4 or DeepSeek V3-0324 without rate limits, and available for a free trial at https://lambda.ai/dwarkesh. To sponsor a future episode, visit https://dwarkesh.com/advertise. 𝐓𝐈𝐌𝐄𝐒𝐓𝐀𝐌𝐏𝐒 00:00:00 – How Llama 4 compares to other models 00:12:20 – Intelligence explosion 00:27:22 – AI friends, therapists & girlfriends 00:35:56 – DeepSeek & China 00:40:35 – Open source AI 00:55:01 – Monetizing AGI 00:59:18 – The role of a CEO 01:02:50 – Is big tech aligning with Trump? 01:07:56 – 100x productivity
Mark Zuckerberg discusses the launch of Llama 4, highlighting its advancements over Llama 3. He emphasizes the growing user base of Meta AI and the importance of personalization in AI interactions. Zuckerberg introduces the new models, Scout and Maverick, and teases the upcoming Behemoth model, which will feature over 2 trillion parameters, aiming to enhance efficiency and user experience.
"Mark, thanks for coming on the podcast again. Yeah, happy to do it. Good to see you. You too. Last time you were here, you had launched Llama 3. Now you've launched Llama 4. Well, the first version..."
Zuckerberg reflects on the competitive landscape of AI, noting the significant progress in open-source models compared to closed-source ones. He discusses the reasoning phenomenon and the specialization of models for specific tasks, asserting that this year marks a pivotal moment for open-source AI as it begins to rival closed-source alternatives.
"I'm also excited about the Behemoth model, which is coming up. It's going to be our first model that's sort of at the frontier — more than 2 trillion parameters. As the name says, it's quite big..."
In this segment, Zuckerberg addresses the challenges of benchmarking AI models, emphasizing the need for metrics that align with user value in Meta AI. He critiques existing benchmarks for being skewed and discusses the importance of focusing on product quality and user feedback rather than just performance metrics.
"negative in others, but overall good — is that it’s not just Llama. There are a lot of good ones out there. I think that's quite good. Then there's the reasoning phenomenon, which you're alluding..."
Zuckerberg envisions a future where AI seamlessly integrates into daily life, enhancing interactions across various platforms. He highlights the potential of full-duplex voice technology and the importance of personalization in creating a compelling user experience, predicting that AI will become a constant companion in our daily routines.
"One of the things we've generally tried to do over the last year is anchor more of our models in our Meta AI product north star use cases. The issue with open source benchmarks, and any given th..."
Zuckerberg discusses the concept of an intelligence explosion, where automating software engineering could lead to rapid advancements in AI capabilities. He shares insights into Meta's efforts to develop coding agents and the implications of AI writing code, suggesting that within 12 to 18 months, AI will significantly contribute to software development.
"However you define it, that's what a lot of the labs are going for. But there's no doubt that different folks have optimized toward different things. I think the Anthropic folks have really focu..."
In this segment, Zuckerberg elaborates on the physical and infrastructural challenges that accompany AI advancements. He emphasizes the importance of building the necessary infrastructure and the co-evolution of AI systems and user interactions, highlighting that both technological and human factors will shape the future of AI.
"I got a chance to play around with both Orion and also the Meta AI app, and the voice mode was super smooth. It was quite impressive. On the point of what the different labs are optimizing for — ..."
Zuckerberg shares insights from Meta's experience with automating ranking experiments in advertising. He explains the limitations faced in testing hypotheses and the need for AI to generate high-quality ideas for testing. This segment underscores the challenges of scaling AI capabilities and the importance of improving the quality of AI-generated hypotheses.
"of different labs doing leading work in different domains. Some will be more enterprise-focused or coding-focused. Some will be more productivity-focused. Some will be more social or entertainme..."
In this segment, Zuckerberg speculates on the future of AI-generated hypotheses and their potential to surpass human capabilities. He discusses the necessary conditions for AI to become a valuable asset in generating ideas and highlights the importance of overcoming real-world constraints before achieving this goal. This forward-looking perspective emphasizes the transformative potential of AI in knowledge work.
"it's not just, “Okay, cool, the thing can write code, and now all of a sudden everything is just improving massively.” There are real-world constraints that need to be overcome. Then you need to..."
Zuckerberg explains why Meta's extensive distribution network positions the company favorably in the AI landscape. He notes the significance of the Meta AI app and its user engagement, particularly in WhatsApp, while addressing the challenges of creating a first-class AI experience in the U.S. This segment highlights the strategic importance of user interaction data in training AI systems.
"So if you buy this view, that this is where intelligence is headed, the reason to be bullish on Meta is obviously that you have all this distribution. You can also use that to learn more things ..."
In this segment, Zuckerberg contemplates the evolving role of AI as a virtual coworker rather than just a question-answering tool. He discusses the potential for AI to assist in complex tasks and the implications for user interaction. This forward-thinking view emphasizes the multifaceted nature of AI's future applications in both work and social contexts.
"different things. Imagine you were sitting at the beginning of the development of the internet and you asked, "What's going to be the main internet thing? Is it going to be knowledge work or mas..."
Zuckerberg addresses the growing phenomenon of people forming relationships with AI, including therapists and friends. He emphasizes the importance of ensuring these interactions remain healthy and valuable. This segment raises critical questions about the societal implications of AI companionship and the need for thoughtful design in AI interactions.
"Look at the evolution of things like Instagram and Facebook. If you go back 10, 15, 20 years ago, it was text. Then we all got phones with cameras, and most of the content became photos. Then the ..."
In this segment, Zuckerberg highlights the disparity between the number of friends people have and their desire for more meaningful connections. He discusses how AI can help fulfill this need for social interaction, particularly for those feeling isolated. This perspective sheds light on the potential of AI to enhance human relationships and address loneliness.
"talking through difficult conversations they need to have with people in their lives. "I'm having this issue with my girlfriend. Help me have this conversation.” Or, "I need to have a hard conve..."
Zuckerberg shares his vision for the future of AI interactions, emphasizing the importance of nonverbal communication and the potential for more immersive experiences. He discusses the development of technologies like Codec Avatars and the need for AI to enhance rather than distract from human interactions. This segment explores the evolving landscape of AI in social contexts.
"So I think a lot of these things — things that today might have a little bit of stigma around them — over time, we'll find the vocabulary as a society to articulate why they are valuable, why th..."
Zuckerberg addresses the competitive landscape of AI infrastructure, particularly in relation to China. He discusses the implications of industrial policies and the need for the U.S. to enhance its data center capabilities. This segment highlights the strategic importance of physical infrastructure in the race for AI dominance and the challenges that lie ahead.
"at the breakfast table and on the periphery of my vision is just a bunch of Reels that are scrolling by. Maybe in the future, my AI girlfriend is on the other side of the screen or something. So ..."
In this segment, Zuckerberg addresses the competitive landscape of AI development, particularly focusing on China's advancements in physical infrastructure and data centers. He discusses the implications of export controls on technology and how they affect the capabilities of AI models like DeepSeek compared to Meta's Llama models.
"It's a real competition. You're seeing industrial policies really play out. China is bringing online more power. Because of that, the US really needs to focus on streamlining the ability to build ..."
Zuckerberg shares insights on the licensing of Meta's Llama models and the importance of open-source AI in the industry. He discusses the balance between encouraging innovation and protecting Meta's interests, emphasizing the need for collaboration with other tech giants while maintaining a competitive edge in AI development.
"So the Llama 4 models, when you compare them to what DeepSeek is doing, are good. I think people will generally prefer to use the Llama 4 models. But there’s this interesting contour where it’s ..."
In this segment, Zuckerberg reflects on the evolution of open-source AI and its implications for the future. He discusses the challenges posed by licensing and the need for companies to engage in meaningful conversations about the use of AI models, while also considering the potential for competition from other open-source initiatives.
"So Sam Altman recently tweeted that OpenAI is going to release an open-source SOTA reasoning model. I think part of the tweet was that they won’t do anything silly, like say you can only use it ..."
Zuckerberg emphasizes the necessity of developing custom AI models tailored to Meta's specific needs. He explains how building proprietary models allows for greater control over performance and functionality, ensuring that Meta can innovate effectively in a rapidly evolving technological landscape.
"Now, if the whole open-source part of the industry evolves in a direction where there are a lot of other great options and the license ends up being a reason why people don’t want to use Llama, ..."
Zuckerberg discusses the significance of open-source models in the AI ecosystem, highlighting how they foster innovation and competition. He raises concerns about the sustainability of open-source initiatives and the potential risks of relying on external models for critical applications.
"have to deal with the model architectures. And they make different size tradeoffs that affect latency and inference cost. At the scale that we operate at, that stuff really matters. We made the L..."
In this segment, Zuckerberg explores the cultural implications of AI models, particularly how they encode values and perspectives based on their training data. He discusses the importance of ensuring that AI systems reflect diverse viewpoints and do not inadvertently perpetuate biases.
"if it's dabbling, or fundamental for them the way that it has been for us. A good example is what's going on with Android. Android started off as the open-source thing. There's not really any op..."
Zuckerberg addresses the security risks associated with using AI models developed in other countries. He highlights the potential for vulnerabilities to be embedded in code, emphasizing the need for caution when integrating external models into critical systems.
"it's important that the standard gets built around American models like Llama. I wanted to understand your logic there. With certain kinds of networks, it is the case that the Apple App Store just..."
Zuckerberg discusses the technique of distillation in AI, which allows for the extraction of intelligence from larger models into smaller, more efficient ones. He highlights the benefits of this approach, including cost-effectiveness and the ability to leverage multiple models for enhanced performance.
"Reasoning, I guess, you could say has values too. But one of the nice things about reasoning models is they're trained on verifiable problems. Do you need to be worried about cultural bias if your..."
In this segment, Zuckerberg addresses the security concerns associated with distilling AI models. He outlines the need for rigorous testing and red teaming to ensure that distilled models do not carry over harmful biases or vulnerabilities, particularly in sensitive applications like coding.
"This is the beauty of distillation. It's one of the things that I think has really emerged as a very powerful technique over the last year, since the last time we sat down. I think it’s worked b..."
Zuckerberg explores various business models for monetizing AI, questioning the sustainability of ad-supported models. He discusses the potential for premium services that cater to high-demand applications, emphasizing the need for diverse revenue streams in the evolving AI landscape.
"That's something I'm pretty confident about and something we've done a lot of research around. But I think this is a very big question. How do you do good distillation? Because there’s so much va..."
Zuckerberg shares insights on his role as CEO in overseeing AI projects at Meta. He discusses the balance between providing direction and empowering teams, as well as the importance of fostering collaboration across different departments to integrate AI effectively into existing products.
"That just speaks to the diversity of different things that need to get created. There are going to be business models at each point along the spectrum. At Meta, for the consumer piece we definit..."
In this segment, Zuckerberg addresses the political landscape surrounding AI governance. He reflects on the importance of maintaining productive relationships with government officials and the need for companies to take ownership of their decisions regarding AI moderation and policy.
"Then there are a bunch of questions like, “do you want the thread for Meta AI in WhatsApp to feel like other WhatsApp threads, or do you want it to feel like other AI chat experiences?” There ar..."
Zuckerberg discusses the evolution of content moderation at Meta, highlighting lessons learned from past approaches. He emphasizes the need for robust systems that empower users and ensure accountability, particularly in the context of AI's growing influence on content management.
"specific models that we design, to have the kind of size and latency parameters that they do. Speaking of politics, there's been this perception that some tech leaders have been aligning with Tru..."
Zuckerberg speculates on the potential for AI to dramatically increase productivity in software development. He discusses the implications of such advancements for creativity and innovation, envisioning a future where AI unlocks new possibilities for human achievement.
"made were in the context of content moderation. It's been an interesting journey over the last 10 years on this. It's obviously been an interesting time in history. There have been novel questio..."
In this segment, Zuckerberg explores the concept of high-leverage work and how it varies week to week. He emphasizes the dynamic nature of the tech industry and the importance of adapting to new challenges and opportunities as they arise.
"a week? What are you doing in that hour? I don't know. Every week is a little bit different. It's probably got to be the case that the most leveraged thing you do in a week is not the same thing ..."
Zuckerberg speculates on the transformative potential of AI in software development, envisioning a future where productivity could increase dramatically. He discusses how this shift could lead to a surge in creativity and cultural pursuits, reshaping society's focus away from basic needs.
"What would be possible if, say, software productivity increased like 100x in two years? What kinds of things could be built that can't be built right now? What kinds of things? That's an interest..."
Zuckerberg predicts that advancements in AI will enhance cultural and social pursuits, making the world richer and more diverse. He reflects on how technology will enable people to express nuanced ideas and connect better through creative media.
"This isn't the 1-2 year thing of what happens when you have a super powerful software engineer. But over time, if everyone has these superhuman tools to create a ton of different stuff, you're g..."
In this segment, Zuckerberg discusses the potential for AI to revolutionize customer support at Meta. He explains how AI could handle a significant portion of customer inquiries, ultimately leading to an increase in hiring for support roles rather than job losses.
"If you think about it, the Meta social media view of the world is that yeah, people are going to spend a lot more time doing that stuff in the future. It's going to be a lot better, and it's goi..."
Zuckerberg shares his approach to seeking advice, emphasizing the importance of having a diverse group of advisors rather than relying on a single individual. He highlights the value of collaboration and working with people who inspire him in a dynamic world.
"seriously been able to contemplate having voice support where someone can just call in. I guess that's maybe one of the artifacts of having a free service. The revenue per person isn't high enou..."