so this week is all about llama 3.1, and we've been talking about open source in general this week because finally we have an open source model that is competitive with any Frontier Model that's out there. Mark Zuckerberg, with the release of llama 3.1, wrote a letter about open source AI being the path forward, and I could not agree more. So we're going to read this letter, I'll break it down for you, and hopefully by the end of it you are also convinced that open source AI is the path forward. Let's read it. So this was published July 23rd, 2024, just a few days ago at the time of recording this video. In the early days of high-performance Computing, the major tech companies of the day each invested heavily in developing their own close Source versions of Unix. Now he's already drawing parallels to technology of the past, which is really the best way to prove your point about how this technology will unfold over the coming decades. It was hard to imagine at the time that any other approach could

develop such advanced software basically only a tight-knit team of developers could build. Software well eventually though open-source Linux gained popularity initially because it allowed developers to modify its code however they wanted and was more affordable. And over time because it became more advanced more secure and had a broader ecosystem supporting more capabilities than any closed Unix so a lot to unpack there he is drawing analogies from open-source Linux and so he starts by saying initially it allowed developers to modify its code basically they put out the open-source project and developers could take it and modify it and make it as close to perfect as they could for their exact use case and it was more affordable obviously if it's open source it's free and the only cost you're really incurring is the development effort to build on top of it then what happened is as more diverse developers from different teams different countries different points of view started to get their hands on this open-source software it actually became more advanced than the closed source version.

and with all of these developers looking at every single line of code, and it is completely transparent. It became hardened and much more secure than closed source could ever hope to be. And best of all, it had a massive ecosystem around it. And that's what Zuckerberg and Meta AI is hoping to do this time around. And this way of thinking didn't just come out of nowhere, and it also wasn't out of the goodness of his heart. He wants to win. He wants to be the dominant platform for the artificial intelligence era. And it all stems from the fact that Facebook lost the mobile Revolution. Basically, they were completely beholden to two major companies, really one: Apple with iOS, and Google with Android. Now, of course, Android is open source, but in terms of who actually made all of the money during the mobile Revolution, it was Apple. And so you had really just a couple companies that were dominating the mobile platform, and the mobile platform was the platform for the social era. And so Meta had this enormous platform risk by being built on top of

these other platforms. They had to pay just to play in their playgrounds, and so Mark Zuckerberg and team are trying to learn from that lesson and get way ahead of the AI curve. Now, they also tried to do the same thing with AR and VR. They were really the first major company to move into and produce a mass market VR AR headset, and the reason is because they thought that was the future computing platform. Whether it actually is is still to be determined. Apple just moved into the space, although the Apple Vision Pro is not really popular. It's way too expensive, too heavy, etc. But for the most part, Meta is really well positioned for the likely platforms of the future AR and VR and AI. So, he goes on to say, I believe that AI will develop in a similar way. Today, several tech companies are developing leading closed models. And he's obviously talking about Anthropic, Google, OpenAI, but open source is quickly closing the gap, and it has closed the gap. Last year, Latu was only comparable to an older generation of

models behind the frontier. And by the way, Frontier basically means the cutting edge, best possible model out there this year. Llama 3 is competitive with the most advanced models and leading in some areas. And Llama 3.1 405B specifically is on par with the best Frontier models out on the market right now. Starting next year, we expect future Llama models to become the most advanced in the industry, but even before that, Llama is already leading in openness, modifiability, and cost efficiency. If AI is going to follow the same pattern as Unix and Linux, we will see that more developers are going to develop on top of open-source AI, thus making it cheaper, better, and more secure. He goes on in this paragraph just to talk about the Llama 3.1 405B release. I've already covered that all this week, so I'm not going to read all of this, but the best part is it is the best choice for fine-tuning and distilling smaller models. It is a very permissive license, basically you could do anything with it.

and the really exciting thing is llama 3.1 405b allows you to generate synthetic data to train any other model you want. So he's not afraid of the competition whatsoever. He's basically saying, Hey, if you want to take our model, make a bunch of synthetic data, and then create your own competitive model, please do. Beyond releasing these models, we're working with a range of companies to grow the broader ecosystem. Amazon, Databricks, and Nvidia are launching full suites of services to support developers fine-tuning and distilling their own models. Now, of these three companies, which is kind of interesting that these are the three listed, Amazon doesn't have their own model, Databricks does, and Databricks is very powerful and actually really good. And I made a test video testing it out, so I'll link that in the description below. And Nvidia just a few weeks ago released NeRF, which is another massive model used to create synthetic data to train other models, and again, very permissive. Then he goes on to say, innovators like Grok and I love

that he not only mentioned Grock in the interview that he just published, a breakdown of, but he's mentioning it here as well. They are one of my favorite companies to work with because the inference speed is just so insane. It unlocks so many different use cases. So innovators like Grock have built low latency, low-cost inference serving for all the new models. The models will be available on major cloud services: AWS, Azure, Google, Oracle, and more. Companies like Scale AI, Dell, Deoe, and others are ready to help enterprises adopt Llama and train custom models with their own data. Now, here is the key sentence, and here is really why they're doing all of this. As the community grows and more companies develop new services, we can collectively make Llama their model, the industry standard, and bring the benefits of AI to everyone. Now, that is why they're doing it. They want their model, their creation, even though it's open source, fully permissive, to be the standard. Because then they get to really set the standard. They get to decide how.

this thing evolves, for the most part, obviously since it's open source, you can fork it, you can do whatever you want with it. But as the company with the most money and the most resources, the most engineering talent, they are most likely going to dictate the future of llama meta. Is committed to open-source AI. I'll outline why I believe open source is the best development stack for you, why open-sourcing llama is good for meta, and why open-source AI is good for the world, and therefore a platform that will be around for the long term. So, I already covered a lot of this in the interview, but I think it's really good to have this all in writing, and we'll cover it right now. So he's going to need to cover quality, he's going to need to cover cost, security, the ecosystem more broadly, how it benefits society, and he's going to do that right now. So he says, When I talk to developers, CEOs, and government officials across the world, I usually hear several themes. We need to train, fine-tune, and distill our

own models. Every organization has different needs that are best met with models of different sizes that are trained or fine-tuned with their specific data. On-device tasks and classification tasks require small models, while more complicated tasks require large models. Now, that is something that I've been talking about a lot in this channel and on my Twitter account for the last few weeks and months. I strongly believe that the future of AI architecture is going to be a lot of small models working together and a lot of the AI compute actually happening on edge devices like desktops, laptops, and mobile phones. That is actually the same approach that Apple intelligence took. The only difference being they're completely closed source. They rolled their own model, they will release it, and it's going to run on device very securely. And then if they need more compute, they will send it off to their private cloud, and then only when absolutely necessary, they are going to call OpenAI's API to get kind of world knowledge questions that these smaller, more narrowly trained models cannot handle. Now you'll be able

to take the most advanced llama models continue training them with your own data, and distill them down to a model of your optimal size without us or anyone else seeing your data. Really, really important, especially for the business use cases. For Enterprise use cases, they're going to want to train the models on their own internal highly sensitive data. They're going to want to roll ATS, self-deploy it themselves, and they don't want anybody else seeing what's going on. Open source is the key to that. And so, if you think about it, would you rather have a very low-cost, highly efficient, very narrowly focused model for your specific use case as a business that you can run on Prem, or would you rather send it off to this larger, generic model that is much more expensive, much slower latency-wise, and also give up all of your data? I think the answer is pretty clear. Next, we need to control our own destiny and not get locked into a closed vendor. Many organizations don't want to depend on models they cannot run and control.

themselves. They don't want closed-model providers to be able to change their model, alter their terms of use, or even stop serving them entirely. They also don't want to get locked into a single cloud that has exclusive rights to a model. Open source enables a broad ecosystem of companies with compatible tool chains that you can move between easily. This is incredibly important as an entrepreneur, as somebody who has built a software company. Platform lock-in and platform risk, in general, is something that really affected my previous company and something I thought about a lot. Now, when I'm consulting large companies, enterprise companies on their AI strategies, I am telling them to build their internal AI tools in a way that abstracts any particular cloud provider, and that is the right strategy to take. Because if you are completely dependent on a single platform, you are basically beholden to any decision that they make about their model. Next, we need to protect our data. Many organizations handle sensitive data that they need to secure and can't send to closed models over cloud APIs. Other organizations

simply don't trust the cloud model providers with their data. Open source addresses these issues by enabling you to run the models wherever you want. It is well accepted that open source software tends to be more secure because it is developed more transparently. Now I have to point out the obvious contradiction here. Facebook specifically was known as the Absolute Data vacuum. They were collecting all of your data. They let other companies use it. If you remember Cambridge Analytica, there were a lot of controversies over the years with Facebook collecting your data and using it in weird ways, to say the least. And so look, obviously they deserve a second chance, and I love the new direction they're taking, open-sourcing it, not trying to own it all. So I appreciate it, but I have to point that out. And everything he said in that paragraph is completely true. There are companies, many companies, who just don't want to give up their data, and maybe even if there are guarantees about data protection, data privacy, they don't trust those companies, and they wouldn't be the

first, we need a model that is efficient and affordable to run. Developers can run inference on llama 3.1 45b on their own infra at roughly 50% of the cost of using closed models like GPT-40 for both user-facing and offline inference tasks. Here's another thing with open source: when all of these developers are looking at it, they are able to find new ways to squeeze out incredible efficiencies from these models. Next, we want to invest in the ecosystem that's going to be the standard for the long term. Lots of people see that open source is advancing at a faster rate than closed models, and they want to build their systems on the architecture that will give them the greatest advantage long term. Now, I want to put this simply: I think it's actually just a function of developers want to build, and when we're able to download the model to play with it ourselves, to build things on top of it, to break it, to rebuild it, I think that's super exciting. But with closed-source models, you can't.

do any of that you simply have an API endpoint. A very structured, very opinionated API endpoint, and you just get what you get. But when you actually own the model, when you can download it on your physical device, that's when things really get interesting. So now, why is open source good for Meta? Meta's business model is about building the best experiences and services for people. To do this, we must ensure that we always have access to the best technology and that we're not locking into a competitor's closed ecosystem where they can restrict what we build. They are specifically calling out Apple here. One of my formative experiences has been building our services constrained by what Apple, yep, okay, will let us build on their platforms. Between the way they tax developers, the arbitrary rules they apply, and all the product innovations they block from shipping, it's clear that many other companies would be freed up to build much better services for people if we could build the best versions of our products and competitors were not able to constrain what we could build.

agree with all of this. Look, Apple's ecosystem is wonderful if you're a consumer, and I know a lot of you will disagree with that. I've used both Android and iOS, and I do have a preference for iOS. I like how clean it is, I like that it just works. But that also means that developers, people who build applications on top of them, are completely beholden to what Apple decides. And look at Epic Games, they've been in a battle with Apple legally for a while now, and I understand why Apple puts in place crazy taxes and rules that are completely arbitrary. So, so I definitely understand that notion. So he is building open ecosystems in AI and AR/VR. People often ask if I'm worried about giving up technical advantage by open-sourcing Llama, but I think this misses the big picture for a few reasons. First, to ensure that we have access to the best technology and aren't locked into a closed ecosystem over the long term, Llama needs to develop into a full ecosystem of tools efficiency.

improvements, silicon optimizations, and other integrations, and they already started publishing some of their ideas. Here it's very rough, and they're actually requesting comment from the developer community about how they want it to be built. Second, I expect AI development to continue to be very competitive, which means that open sourcing any given model isn't giving away a massive advantage over the next best model at that point in time. Yes, agreed, the path for llama to become the industry standard is by being consistently competitive, efficient, and open generation after generation. Third, a key difference between meta and closed model providers is that selling access to AI models isn't our business model. Now, that is kind of a bad business model to be in, is selling access to one particular model because all of the reasons listed in this article here. As a company, if you're buying inference, you are locked into that one model, and if you have different needs over time, it's really difficult to switch. And selling inference to a single model is also, as I said, a really bad business to be in.

because models are becoming commodities, intelligence is becoming a commodity, and I love saying that. That means openly releasing llama doesn't undercut our revenue, sustainability, or ability to invest in research like it does for closed providers. This is one reason several closed providers consistently lobby governments against open source. They say it's from a position of AI security, but it's not. It's simply regulatory capture. Finally, Meta has a long history of open source projects and successes. So, I actually didn't know how big Meta's history was of open source projects, but it is true. They have the Open Compute Project, PyTorch, React, which is basically the go-to front end that many developers choose when building new applications, and many more tools. In a previous video, I listed all the tools. It's like dozens if not hundreds. So, why is open source AI of the world? It is necessary for a positive AI future. AI has more potential than any other modern technology to increase human productivity, creativity, and quality of life, to accelerate economic growth while unlocking progress in medical and

scientific research. If you're watching this channel, I am so happy that you're here because you get to experience the birth of this technology. I am 100% convinced will change the world completely. I think it has the potential to be 10 times bigger than the internet in terms of impact on the economy, on productivity, everything that I just mentioned. Open source will ensure that more people around the world will have access to the benefits and opportunities of AI, that power isn't concentrated in the hands of a small number of companies, and that the technology can be deployed more evenly and safely across society. Now he's going to go on to talk about safety, and a lot of people say closed source is safer than open source. He says, I think governments will conclude it's in their interest to support open source because it will make the world more prosperous and safer. My framework for understanding safety is that we need to protect against two categories of harm: unintentional and intentional. Now when he says unintentional, he means an AI

system may cause harm even when it was not the intent of those running it to do so. Rogue AI, so for example, modern AI models may inadvertently give bad health advice. Or in more futuristic scenarios, some worry that models may intentionally self-replicate or hyper-optimize goals to the detriment of humanity. That is the paperclip example. So for those of you who aren't familiar, I believe it goes, if AI is optimized to make the most amount of paper clips, they will do anything to do so, including killing off all humans. Intentional harm is when a bad actor uses an AI model with the goal of causing harm. So unintentional harm covers the majority of concerns people have around AI, ranging from what influence AI systems will have on the billions of people who will use them to most of what truly catastrophic science fiction scenarios for humanity. On this front, open source should be significantly safer, since these systems are more transparent and can be widely scrutinized. Agreed, historically, open source software has been more secure.

this reason similarly using llama with its safety systems like llama guard will likely be safer and more secure than closed models. And I agree when you have vastly more developers and engineers and research scientists looking at the code looking at the architecture then of course it's going to be more scrutinized because it's more transparent and thus it should be more secure and more safe. So he goes on to talk about their safety process rigorous testing red teaming and since the models are open anyone is capable of testing them for themselves. And we must keep in mind that these models are trained by information that's already on the internet so the starting point when considering harm should be whether a model can facilitate more harm than the information that can quickly be retrieved from Google or other results. When reasoning about intentional harm it's helpful to distinguish between what individual or small scale actors may be able to do as opposed to what large scale actors like nation states with vast resources may be be able to do. Individual Bad actors in the future may

be able to use the intelligence of AI models to fabricate entirely new harms from the information available on the internet. At this point, the balance of power will be critical to AI safety. I think it would be better to live in a world where AI is widely adopted so that larger actors can check the power of smaller bad actors. So one AI system versus another AI system, this is how we've managed security on our social networks. Our more robust AI systems identify and stop threats from less sophisticated actors who often use smaller scale AI systems. More broadly, larger institutions deploying AI at scale will promote security and stability across society. So the next question is how the US and democratic nations should handle the threat of states with massive resources like China. Now, if you're thinking the same as I'm thinking, this kind of reminds us of the situational awareness paper that we covered a few weeks ago where Leupold Ashenbrener talked a lot about China in particular stealing our AI secrets.

Leopold was very much about private enterprise and government working together to protect these secrets. Mark Zuckerberg and The Meta team are taking a different approach, same thought of we need to protect it, but if everybody has it, then we're safer. The United States advantage is decentralized and open innovation. Some people argue that we must close our models to prevent China from getting access to them, but my view is that this will not work and will only disadvantage the US and our allies. Our adversaries are great at espionage; they have proven this, stealing models that fit on a thumb drive is relatively easy, and most tech companies are far from operating in a way that will make this difficult. It seems most likely that a world of only closed models results in a small number of big companies, plus our geopolitical adversaries having access to leading models. Yeah, basically what he's saying is if only a handful of closed source companies have the most important secrets in the whole world and China has already proven their capability to hack and perform espionage.

and steal these secrets. Why are we to think that that's not going to happen again? And really, it only takes one time. It only takes one mistake, kind of like how CrowdStrike deployed something that completely took down half of the world's computers. It only takes one mistake like that, and then all of a sudden we lose all of our secrets. It seems most likely that a world of only closed models results in a smaller number of big companies, plus our geopolitical adversaries having access to leading models, while startups, universities, and small businesses miss out on opportunities. Plus, constraining American innovation to closed development increases the chance that we don't lead at all. Completely agreed. Our best strategy is to build a robust open ecosystem and have our leading companies work closely with our government and allies to ensure they can take best advantage of the latest advances and achieve sustainable first-mover advantage over the long term. The bottom line is that open-source AI represents the world's best shot at harnessing this technology to create the greatest

Economic opportunity and security for everyone. And I agree. And if you followed this channel for any period of time, you know that I agree. You know that I love open source and all of this stuff makes sense. But if I'm being completely honest, the reason I love open source is 'cause I like to tinker with things. I want to download it on my computer. I want to play around with it. I want to know that I own it. And that's incredibly important. And I haven't always been this way. I just admitted that I like Apple's ecosystem for my iPhone. But in this case, there's something about being able to download artificial intelligence to my laptop, to my phone, to my desktop, play around with it, break it, train it the way I want to, turn off my internet, and still have access to the world world's knowledge. That just makes me excited about the future. If you enjoyed this video, please consider giving a like and subscribe, and I'll see you in the next one.

Zuckerberg's Bold Vision: Open-Source AI Leading the Future

Artificial Intelligence (AI), Open Source, Generative Art, AI Art, Futurism, ChatGPT, Large Language Models (LLM), Machine Learning, Technology, Coding, Tutorials, AI News, and more

** Exclusive Pinecone Partner **