Meta Platforms Just Launched A Potentially Dangerous AI Game

The race to build the best AI models is certainly on, with OpenAI, the current leader, emerging in late 2022, and rising competitors, the alphabet‘s (Google -0.28%) (Google -0.17%) Gemini and AI startup Anthropic are also building high-performance models with hundreds of billions of parameters.

However, investors should not rule out Meta platforms (Meta 2.71%)which takes the most unique and distinctive approach to building generative AI models.

Will this approach allow Meta to get ahead of the industry, or do the risks outweigh the benefits?

When competitors close their doors, Meta opens.

Last week, Mark Zuckerberg and Meta celebrated the release of Meta’s Llama 3.1 model, which includes 405 billion parameters.

This is a massive model. So far, Meta has only released 8 billion parameters and 70 billion Llama models. But Llama 3.1 405B is Meta’s first shot at “frontier” models, pushing the boundaries of the most advanced large language models for generative AI on the planet. According to Meta, the new model outperforms even the most advanced large language models, like ChatGPT 4-Omni and Anthropic’s Claude 3.5 Sonnet model, on many — but not all — parameters. Zuckerberg also claims that Llama 3.1 inference costs just 50% of what ChatGPT 4o would run.

Not only does Llama’s performance match or outperform many of the best models on the market today, but it also has the distinction of being the only open source boundary model.

Advantages of open source

Open source software means that the license holder allows third parties to freely access and modify the source code of the software. External developers can then make changes to the software, allowing them to improve functionality, fix bugs, or improve security.

By distributing the software code for free, developers have the opportunity to improve the product more quickly than with a “closed” software system where only company employees have access to the code to modify it. Since outside developers are drawn to “free” open source software rather than expensive proprietary software, the open source model is preferable if one wants to scale up usage quickly. In addition to these general advantages of open source, Llama can be run anywhere, so developers do not have to send their private data to a closed model or a specific cloud.

That’s why Mark Zuckerberg believes Meta AI will become the world’s most widely used AI assistant by the end of this year, overtaking OpenAI.

in Blog post In a recent press release, Mark Zuckerberg wrote that he sees the development of LLM programs as progressing in a similar fashion to the development of Linux and UNIX operating systems in the 1990s. While the closed UNIX operating system was an early pioneer, the advantages of Linux’s open-source model eventually paved the way for it to become the standard operating system for cloud computing and mobile devices.

Zuckerberg also points out that Meta has benefited from open source tools in the past, such as its development of open source data center architecture and AI software like Pytorch, which Meta initially developed. Since Meta’s core business is advertising, not monetizing software or data center architecture directly, open source hasn’t hindered its ability to generate revenue and invest in R&D. At the same time, Meta has ultimately saved billions of dollars by letting these tools be developed with the help of third parties and then using them to power its core social media platforms.

The same goes for Llama AI. Meta isn’t a public cloud software or services company, so it doesn’t really rely on selling its model directly to generate revenue.

But Meta has to take advantage of the llama… eventually.

But that doesn’t mean Meta doesn’t eventually expect to make money from Llama. It’s more likely to do so by using the core model and building services on top of it. In a conference call with analysts in Q1, Zuckerberg hinted at building AI systems to serve customers via WhatsApp business messages, injecting ads into Meta AI’s AI interactions, or perhaps charging for access to larger AI workloads with more compute.

Other open source software companies, such as Red Hat, have already exploited AI models by selling customer services and consulting. WordPress has also used a dual-licensing model to monetize its website-building software, offering large companies a paid “deluxe” version. So there is precedent for leveraging open source models.

Meta is unlikely to generate any direct revenue from Llama in the near term, but over the next few years, the company should begin rolling out and leveraging additional services built on top of the core Llama model.

Defects

But there are also risks to open source – especially In the context of artificial intelligence. These risks relate to privacy and security, two elements that Meta has been criticized for in the past. Concerns about privacy and security are likely the reason behind the talks with apple (AAPL 0.22%) An attempt to put Llama on Apple’s new Intelligence platform failed earlier this year. Apple has perhaps the most closed business model in tech, and has always placed a high value on privacy and security. So the Meta spat and exclusion from Apple Intelligence could be a warning sign.

Specifically, there may be concerns about making the source code and model weights publicly available, as this could give bad actors the opportunity to modify the code for malicious purposes. This would expose them to interference from government authorities, who may restrict the distribution of the Lamma program to prevent it from falling into the hands of hostile governments, such as China, Russia, or Iran.

Zuckerberg’s response is that these countries are adept at spying and would likely be able to access leading edge models anyway, even in a closed system. This essentially means that fewer “good guys” will have access to leading edge models (only a few big companies), which actually risks putting the US at a disadvantage. Additionally, Zuckerberg believes that stifling open innovation risks the US and its allies never having the best AI.

But this argument seems vague: since these bad guys will have access to leading AI models anyway, should we give them the code for free? It’s unclear whether this argument will actually sway the US government to go along with this line of thinking.

A big factor in the AI ​​race

Meta’s choice of open-source AI software could make it the world’s leading AI company, and thus the world’s most valuable one day. But this approach also risks government oversight and regulation, threatening its ability to compete in the AI ​​wars at all.

For Meta investors and AI investors in general, the rollout of Llama 3.1 and the potential government reaction to it during an election year are key points to watch in the coming months.