Against all odds, LLaMa, the artificial intelligence model proposed by Meta, is experiencing remarkable success. It risks overshadowing ChatGPT, Google Gemini, and other stars in the field. How can we explain this rapid and sudden rise in popularity?
LLaMa, the Crusader of Open Source
What characterizes LLaMa is that it is an open source LLM. This means it is accessible for free, but also that anyone can intervene in the code and refine certain aspects. An open-source project can benefit from the contributions of thousands of top-notch developers. This is one of the aspects that made Linux popular: security was enhanced by armies of vigilant contributors.
For the user, an advantage is that it is possible to download it entirely to one’s computer – although a powerful setup is required for the most advanced version – and thus use it without an Internet connection. One just needs to be aware that its knowledge is limited to December 2023.
Nevertheless, LLaMa is an LLM produced by Meta, the company that manages Facebook, Instagram, and WhatsApp. Hundreds of millions of dollars have been invested in its development.
Why did Zuckerberg choose such an approach? We will come back to this.
Acclaimed Performance
The open-source factor would be negligible if the performance wasn’t there. Yet, with each version, LLaMa has increased its capabilities. It’s also true that the versions followed one another quickly:
- LLaMa 1: February 2023
- LLaMa 2: July 2023
- LLaMa 3: April 2023 – available in two versions: 8B (8 billion parameters) and 70B (70 billion).
- LLaMa 3.1: late July 2024
It was from version 3 that LLaMa began to receive strong praise, with some even calling it “superpowerful”. Its performance has been recognized by benchmarks like MMLu. The success rates of the three main LLMs were as follows:
| Gemini Ultra | 90 % |
|---|---|
| GPT-4 | 86 % |
| LLaMa 70B | 82 % |
LLaMa 3.1 represented a considerable step forward, as the 405B version relies on 405 billion parameters. This time, in the MMLu test, the score obtained (88.6%) is just slightly lower than that of GPT-4.0 (88.7%).
Why Did Meta Choose the Open Source Formula?
Mark Zuckerberg took advantage of LLaMa to establish himself as a supporter of the open-source formula. In reality, Meta benefits in more than one way.
- In Silicon Valley, Sun Tzu’s principles are appreciated, and one of them is: “if you can’t win on a battlefield, change the battlefield.” By offering LLaMa for free, Meta undercuts its three main competitors: OpenAI, Google, and Anthropic (Claude). This, before they could capture the bulk of the AI market.
- Meanwhile, Meta can restore its image following the privacy negligence issues that tarnished its history.
- Due to the nature of open source, LLaMa can benefit from the feedback of thousands of developers.
- The nature of open source and non-proprietary nature can reassure many companies worried that AIs like ChatGPT or Gemini might analyze their information. Thus, Samsung, Amazon, and Apple banned the use of ChatGPT as early as the end of spring 2023. The same goes for various law firms, hospitals, or government organizations.
- Meta is capable of offering integrating LLaMa 3 into companies for custom tasks while boasting proven technology.
- Where Amazon, Microsoft, and Google have dominated the cloud, Meta finds in AI a field where it can establish itself on a large scale against other GAFAM.
- Zuckerberg is now interviewed as an open-source guru and can make statements such as “open-source AI will become the industry standard.” ” LLaMa is not a technology but an ecosystem“… In fact, LLaMa 3 was quickly integrated into the offerings of giants like Amazon Web Services and Microsoft Azure, Zoom, and AT&T.
In Plain Language?
To reach the general public, an interface as simple to use as ChatGPT is required. It exists in the form of an app named Meta AI. Just like ChatGPT, it can generate images and even animate them.
However, what may have hindered Meta AI’s audience is that, as of early September 2024, this app was not yet accessible in Europe (officially due to regulatory issues) and therefore in France.
