Meta’s Llama Revenue Plan, Anthropic Partners with Brave, & 1X’s Neo Gamma Begins Testing

Summary

AI Deep Dive Podcast Summary Episode: Meta’s Llama Revenue Plan, Anthropic Partners with Brave, & 1X’s Neo Gamma Begins Testing
Release Date: March 22, 2025
Host: Daily Deep Dives

The latest episode of the AI Deep Dive podcast, hosted by Daily Deep Dives, delves into significant developments shaping the artificial intelligence landscape. The discussion centers on Meta’s monetization strategies for their Llama models, Anthropic’s partnership with Brave Search, the rollout of 1X’s Neo Gamma humanoid robots, and Microsoft’s pioneering research on data ownership. Hosts Alex and Blake provide in-depth analysis, enriched with insightful quotes and expert perspectives.

---

1. Meta’s Llama Revenue Plan

The episode opens with an exploration of Meta’s approach to monetizing their open-source Llama AI models. Initially released with a focus on democratizing AI, recent revelations in the Cadre v. Meta lawsuit indicate a more intricate financial strategy.

Key Points:

• Monetization Without Direct Charges: Meta offers Llama models for free, but capitalizes by sharing revenue with host partners providing infrastructure and cloud services.
• Strategic Partnerships: Major companies like AWS, Nvidia, Databricks, Dell, Azure, Google Cloud, and Snowflake are listed as Llama host partners, facilitating easier access and use for developers.
• Future Financial Projections: Meta plans to double its AI investment to $60-$80 billion in 2025, hinting at future revenue streams such as subscriptions and licensing.

Notable Quotes:

• Alex [00:07]: “It's not just about the tech itself... it's also about how businesses are figuring out how to make money from it.”
• Blake [02:28]: “Meta has actually listed a whole bunch of companies as Llama host partners. And these are big names.”

Insights: Meta’s strategy leverages the convenience factor for developers, akin to offering a “cake mix” instead of individual baking ingredients, thereby embedding their AI deeper into various platforms without directly charging for the models. This indirect monetization approach not only facilitates widespread adoption but also mitigates legal risks by fostering a community around Llama.

---

2. Anthropic’s Partnership with Brave Search

Shifting focus, Alex and Blake discuss Anthropic’s integration of Brave Search into their Claude chatbot, unveiling behind-the-scenes collaborations that enhance AI functionalities.

Key Points:

• Use of Brave Search: Contrary to expectations of developing an in-house search engine, Anthropic utilizes Brave Search, a privacy-focused platform, for web searching capabilities in Claude.
• Evidence of Partnership: Indicators such as shared citations and the presence of “bravesearch Params” in Claude’s code confirm the collaboration.
• Industry Implications: This partnership highlights a trend where AI companies opt for existing specialized tools rather than building from scratch, promoting efficiency and leveraging niche expertise.

Notable Quotes:

• Blake [04:25]: “Anthropic recently added Brave Search to their list of sub processors.”
• Alex [05:06]: “Brave. And is it the only time Brave has been involved with AI? Right?”

Insights: Anthropic’s decision to partner with Brave Search underscores the importance of privacy and specialized technology in AI development. It also signifies a shift towards more transparent collaborations in the AI industry, contrasting with competitors like OpenAI, which maintain more secrecy around their search integrations.

---

3. 1X’s Neo Gamma Begins Testing

The conversation transitions to the tangible integration of AI through humanoid robots, focusing on 1X’s Neo Gamma and its upcoming home testing phase.

Key Points:

• Early Testing Phase: 1X aims to deploy Neo Gamma in hundreds to thousands of homes by the end of 2025, allowing real-world interaction and data collection.
• Human-Assisted Operation: Initially, Neo Gamma operates under remote human control to ensure safety and gather accurate environmental data.
• Functional Capabilities and Challenges: Demonstrations showcased Neo Gamma performing basic chores, though issues like hardware malfunctions highlight the complexities of reliable home robotics.
• Privacy Considerations: The robot’s sensors and cameras raise significant privacy concerns, with 1X offering users control over sensor access by employees.

Notable Quotes:

• Blake [06:17]: “Imagine AI in physical forms interacting with our everyday world.”
• Alex [07:17]: “So basically, humans are controlling the robots remotely until the AI is good enough to do it on its own.”

Insights: 1X’s Neo Gamma represents a pivotal step towards integrating AI into daily life, balancing innovation with necessary precautions. The remote control mechanism serves as a safeguard during the developmental phase, ensuring that real-world data is collected responsibly while addressing potential privacy issues.

---

4. Microsoft’s Research on Data Ownership and Data Dignity

Concluding the episode, Alex and Blake delve into Microsoft’s groundbreaking research on data ownership, a critical issue amidst rising legal challenges surrounding AI training data.

Key Points:

• Training Time Provenance: Microsoft is developing methods to trace and credit the original sources of data used in AI training, aiming to recognize and compensate creators.
• Impact of Legal Landscape: With lawsuits alleging unauthorized use of copyrighted material, Microsoft’s initiative addresses the ethical and legal implications of AI data utilization.
• Data Dignity Concept: Advocate Jaron Lanier emphasizes the importance of connecting AI outputs to original creators, promoting fair compensation and recognition.

Notable Quotes:

• Alex [09:08]: “Microsoft is calling this research training time provenance. Basically, it's about figuring out how to trace back the influence of specific data used to train an AI.”
• Blake [10:22]: “Imagine, he says, that an AI creates an image that looks a lot like a specific artist's style. That artist or their family could be acknowledged, even compensated.”

Insights: Microsoft’s efforts signify a potential paradigm shift towards more ethical AI development, where the contributions of data creators are acknowledged and rewarded. This contrasts with industry norms, where data usage often lacks transparency and direct compensation, highlighting a move towards more sustainable and respectful AI practices.

---

Conclusion: Balancing Innovation with Ethical Considerations

Throughout the episode, Alex and Blake weave together narratives of technological advancement and the underlying business and ethical frameworks shaping AI’s future. From Meta’s nuanced revenue strategies and Anthropic’s strategic partnerships to 1X’s pioneering robotics and Microsoft’s data ownership research, the discussion underscores the multifaceted nature of AI development.

Final Reflections: The hosts pose a fundamental question to listeners: As AI becomes more integrated into our lives, how do we balance its innovative potential with the protection of individual rights and contributions? This prompts reflection on the evolving roles of creators, developers, and users in an increasingly AI-driven world.

Notable Closing Quote:

• Alex [12:09]: “We've seen how the business side of AI is taking shape... But under it all is the question of data ownership and fair compensation for creators.”

This episode of AI Deep Dive offers a comprehensive overview of current AI trends, thoughtfully examining both the opportunities and challenges that lie ahead. Whether you’re a tech enthusiast, developer, or simply curious about AI’s trajectory, the insights provided ensure you remain informed and engaged with the evolving AI landscape.

Transcript

[Alex]

So it's becoming pretty clear, isn't it, this whole AI thing. It's not just about like the tech itself, you know, it's also about how businesses are, well, figuring out how to make money from it. And even more interesting, I think is the whole question of like, who actually owns the data that makes all this AI stuff possible in the first place. And that's what we're going to dive into today.

[Blake]

Yeah, for sure. There's been a lot happening lately, you know, just a bunch of developments that when you take a step back and look at them all together, it really gives you a sense of where things are headed. So we're going to be digging into how Meta is approaching the whole monetization thing with their Llama models. Then there's this whole partnership that seems to be going on with Anthropic and their Claude search, which is pretty interesting. And we'll even touch on like the future of humanoid robots in our homes and some really forward thinking research that Microsoft is doing on data ownership.

[Alex]

Yeah, let's start with Meta and Llama. If you remember, when those open source models came out, the whole narrative was about like democratizing AI, right? Not really about making a buck directly. I mean, even Zuckerberg kind of hinted at that. But then recently there was this unredacted court filing in the Cadre v. Meta lawsuit and it, well, it kind of changes things.

[Blake]

Right? So think about the contrast here. We have this whole public image of open access and all that, but then behind the scenes there's this whole financial thing going on and this unredacted filing. Which means, by the way, that parts of this legal document that were hidden from us are now public. Well, it shows how Meta actually shares a percentage of the revenue with the companies that host those Llama models for users.

[Alex]

Okay, so I get it. The core Llama models are still free, right? But the companies that provide like the infrastructure, the cloud hosting and all those extra services on top of them, they're making money and Meta is getting a cut. It's actually a pretty clever way to like monetize without actually charging for the models themselves.

[Blake]

Exactly. And to connect this back to the bigger picture for you, Meta has actually listed a whole bunch of companies as Llama host partners. And these are big names. Aws, Nvidia, Databricks, Grok, Dell, Azure, Google Cloud, even Snowflake. Now the court filing makes it clear that developers can still download and run the models on their own if they want to. But you know, these hosting providers, they make everything so Much easier for developers. They provide all the tools and services you need to just get started.

[Alex]

It's kind of like, you know, you could buy flour, sugar and eggs and bake a cake from scratch, or you could just buy a cake mix. Both get you a cake. But one is way easier. And it seems like Meta is tapping into that convenience factor with these AI. What did you call it?

[Blake]

AI Cake Mix providers.

[Alex]

Right. AI Cake mix providers. Makes sense. But you know, it's interesting because back in April of last year, Zuckerberg actually mentioned during an earnings call that they might try to make money from Llama through things like licensing, business messaging, integrations, and even ads within AI interactions.

[Blake]

So why the change? Well, more recently, Zuckerberg has been saying that the real benefit for Meta is all the improvements to LLAMA that come from the open source community and how well it integrates into their own products like the Meta AI Assistant. He even said that this open approach is good business for them because it makes their own stuff better.

[Alex]

So win, win, right? The community benefits, Meta gets better products. Sounds good to me. But here's the thing. Meta is planning to spend a lot more money on AI this year. Projections are somewhere between 60 and 80 billion dollars in 2025. That's like double what they spent in 2024. So how are they going to make that money back? And then there were those rumors about them thinking about a subscription service for Meta AI with like extra features and stuff.

[Blake]

So here's what it all means for you. It looks like Meta has a multifaceted strategy for making money from their AI work. Revenue sharing with those host partners is a direct way, even if it's not as well known. And if we remember that copyright lawsuit, the Cadre v. Meta case, it makes you think, right? Could this approach be a way for Meta to deal with those legal issues? Maybe by creating this big community around Llama and making money indirectly, they're trying to like, lessen the risks that come with how the model was trained.

[Alex]

That's a really interesting point. Okay, let's switch gears a bit. Anthropic's Claude Chatbot can now search the web just like some of its competitors. But the big question is what tech are they using to do that?

[Blake]

Right. You might think that a company like Anthropic would build their own search engine, but all the evidence suggests that they're using, get this, Brave Search.

[Alex]

Brave, the privacy focused browser. That's kind of unexpected. What makes you think so?

[Blake]

Well, first, Anthropic recently added Brave Search to their list of sub processors. Basically, those are third party companies that Anthropic uses for certain tasks, like in this case, data for Claude. Second, people have seen the same citations in Both Claude's and BraveSearch's results for the same searches. And lastly, and this is probably the most convincing piece, a software engineer found something called bravesearch Params in the code for Claude's web search.

[Alex]

Brave Search Params. Yeah, that sounds pretty clear. Brave. And is it the only time Brave has been involved with AI? Right. I think I read somewhere that they also power the search for Mistral's lechatbot. It seems like Brave is becoming a big part of the AI world behind the scenes.

[Blake]

It really does. And it's important for you to know that some AI companies are very secretive about who they partner with for search, probably because of competition. OpenAI, for example, works with Bing, but they're also thought to use other secret sources for ChatGPT's search. So anthropic being more open, or at least the evidence suggesting they are, is pretty interesting.

[Alex]

It makes you wonder why they went with Brave. Right? Maybe they both care about privacy. Or maybe Brave has some special tech that other search engines don't.

[Blake]

Good question. It could be shared values, some special tech, or maybe just a good business deal. We don't know for sure, but it does show you something important about the AI world. Companies are using existing tools and expertise instead of always trying to build everything themselves. Makes sense, right?

[Alex]

Totally. Okay, ready for a big change of pace? We're going from virtual AI to, well, physical robots. Specifically the Norwegian startup 1X and their humanoid robot Neo Gamma.

[Blake]

Now this is where AI gets really real for you. Imagine AI in physical forms interacting with our everyday world. 1X wants to start testing NeoGamma in a few hundred, maybe even a few thousand homes by the end of next year. 2025. Their CEO Bernd Bernick talked about getting early adopters involved by letting the robot live and learn among people.

[Alex]

Yeah, humanoid robots for the home are getting a lot of attention lately. We've got companies like Figure getting huge valuations and there are even rumors about OpenAI getting into it. But actually putting robots in people's homes, that feels like a big step. Kind of like the early days of self driving cars. So much potential, but also some risks.

[Blake]

Absolutely. And Bernich is pretty upfront about the fact that Neo Gamma is still in its early stages. It's not fully autonomous or ready for everyone to buy just yet. For these early tests, they're doing what they call bootstrapping that means remote human operators will be watching the data from the robot's sensors and cameras and controlling it in real time.

[Alex]

So basically, humans are controlling the robots remotely until the AI is good enough to do it on its own. That way they can collect real world data without the robot, you know, going rogue.

[Blake]

Precisely. They're gathering information on how the robot handles a real home environment to improve and train their AI models. Smart, right?

[Alex]

Makes sense. But you got to think about privacy too. These robots have microphones and cameras, and they're going to be in people's homes.

[Blake]

Yeah, that's a huge concern for you. 1x says that users will be able to Decide when a 1x employee can access the robot sensors for things like checking up on it or controlling it remotely. But having this kind of tech in our homes, that's something we'll need to think about carefully as it becomes more common.

[Alex]

So what can neogamma actually do right now? I saw some footage from that demo they did@Nvidia GTC.

[Blake]

Oh yeah, in that demo, they showed neogamma doing basic chores like vacuuming and watering plants. It can move around a room without bumping into things, but it wasn't perfect. The robot started shaking and it just fell over. They said it's because of bad WI fi and a low battery. Shows you how hard it is to get these complex robots to work reliably in the real world.

[Alex]

Yeah, definitely. Still early days. And to be honest, the details about this early adopter program are still a bit vague. They have a wait list on their website, but we don't really know what it'll be like for those early users, especially with the whole remote control thing.

[Blake]

Right. We don't have all the answers yet. It seems like a small group of people might get to try out this early human assisted version of neogamma. But robots that are totally independent and ready for everyone to buy, that's still a ways off. Still, these early tests are important. You know, they're getting real world data that will help them build better robots in the future.

[Alex]

And speaking of AI training data, that brings us to our last topic. Microsoft is researching how to actually credit the people whose data is used to train these AI models. This seems really important, especially with all the lawsuits about copyright and AI.

[Blake]

It is. Microsoft is calling this research training time provenance. Basically, it's about figuring out how to trace back the influence of specific data used to train an AI. They want to be able to estimate the impact of individual pieces of data, like a picture, some text, or whatever. On what the AI produces.

[Alex]

And get this, their reasoning for doing this research. They laid it out in a job listing that popped up again recently is all about giving incentives, recognition, maybe even payment to the people who provide data for these AI. AI models. That's a big change from how things are done now.

[Blake]

It is. Right now a lot of these neural networks are like black boxes. We don't really know where their outputs come from. But as that job listing says, there are good reasons to change this. And this is all happening while there are tons of lawsuits about intellectual property. Even Microsoft is being sued by the New York Times and some software developers who say their copyrighted stuff was used to train AI without permission.

[Alex]

Right. And Jaron Lanier, a researcher at Microsoft, has been talking a lot about data dignity. It's this idea that we should be able to connect digital creations back to the people who made them and maybe even pay them when their work helps create valuable AI outputs.

[Blake]

Yeah. Lanier wrote a great piece in the New Yorker in 2023 laying it all out. Imagine, he says, that an AI creates an image that looks a lot like a specific artist's style. That artist or their family could be acknowledged, even compensated. There are already companies like Bria, Adobe and Shutterstock trying things like this, but the big AI labs haven't really adopted it yet.

[Alex]

Yeah, because most of the big players like Google and OpenAI, they mostly offer opt out options for copyright holders. You can ask them not to use your stuff, but it's not always easy. And it doesn't fix the problem of data that's already been used.

[Blake]

So you can see Microsoft's research in a few different ways. Maybe it's a genuine attempt to make AI development more ethical and sustainable. Or maybe it's just a test, something they're trying out but might not actually use. There's even the possibility that it's what some call ethics washing making themselves look good while maybe not really changing anything.

[Alex]

Especially when you consider that Those same big AI labs, including Google and OpenAI, have also been pushing for weaker copyright laws when it comes to AI training. They want a really broad definition of fair use. Kind of contradictory.

[Blake]

It is. So Microsoft doing this research is definitely interesting, but we don't know if it means the industry is really going to change how it deals with data ownership and paying creators. It's definitely a different approach compared to what some of their competitors are doing.

[Alex]

All right, let's wrap this up. We've seen how the business side of AI is taking shape, sometimes in surprising ways, like revenue sharing through cloud partnerships. We've also looked at the complex and sometimes hidden infrastructure that powers AI, like Anthropic using Brave Search. And then there's the very real progress and the challenges of bringing AI into our world with those humanoid robot tests. But under it all is the question of data ownership and fair compensation for creators.

[Blake]

So this is the big question for you. As AI becomes more and more a part of our lives, how do we balance all this amazing potential for innovation with protecting the rights and contributions of individuals? Thinking about all these different approaches to making money, building AI systems and dealing with data ownership, what kind of AI future do you see, and what role do you think individual creators and users should have in it?

[Alex]

Yeah, definitely gives you a lot to think about. Thanks for joining us for this deep dive.