show index hide index
|
IN BRIEF
|
Meta recently introduced Llama 3.1, an open source language model that promises to redefine the standards of artificial intelligence. By surpassing the performance of the highly acclaimed GPT-4, this new model offers a series of innovations that could transform the way we interact with language technologies. With better contextual understanding, smoother text production and accessibility for developers, Llama 3.1 marks a major turning point in the field of computational linguistics and demonstrates Meta’s growing ambitions in the race for advanced AI.
Meta launches Llama 3.1: towards a new era of open source AI
Meta has just lifted the veil on Llama 3.1, its latest creation in terms of open source language models. With more than 400 billion parameters, Llama 3.1 poses as a strong competitor against industry behemoths such as GPT-4 and Claude 3.5. This announcement marks a crucial step for innovation in open source artificial intelligence.
Technical characteristics of Llama 3.1
Version 3.1 of Llama is undoubtedly the most powerful ever developed by Meta. This model encompasses 405 billion parameters and has been trained on more than 16,000 Nvidia H100 GPUs, a feat of modern technology. Compared to previous versions 8B and 70B, Llama 3.1 represents a major step forward in complexity and capabilities.
Performance and benchmarks
Meta claims that its new model outperforms OpenAI GPT-4 and Anthropic Claude 3.5 on several benchmarks. This positions Llama 3.1 as a viable alternative to proprietary solutions, with a production cost just a fraction of that of GPT-4.
Collaboration and open source
To encourage adoption and continued development of Llama 3.1, Meta is partnering with giants like Microsoft, Amazon, Google and Nvidia. This collaboration aims to expand the open source ecosystem, similar to what Linux has achieved in the operating system space.
Application and accessibility
Llama 3.1 will be integrated into various Meta applications, including Instagram, Facebook and WhatsApp. It will also be accessible via Meta Quest VR headsets, and will support new languages such as French, Spanish, German and Italian. For those who want to test this AI now, a VPN can be used to fake a connection from the United States.
Comparison table between Llama 3.1 and GPT-4
| Criteria | Llama 3.1 | GPT-4 |
| Number of parameters | 405 billion | 175 billion |
| Performance on benchmarks | Superior | Less efficient according to Meta |
| GPUs used for training | 16,000 Nvidia H100 | Not specified |
| Development cost | Cheaper | Very high |
| Accessibility | Open source | Owner |
| Collaboration with companies | Microsoft, Amazon, Google, Nvidia | Limited |
| Planned adoption | Extended to multiple apps | Limited to certain uses |
| Languages available | French, Spanish, German, Italian | Multilingual |
Opinions and future outlook
According to Mark Zuckerberg, the development of open source AI will lead to rapid growth and mass adoption by developers. Meta hopes that Llama 3.1 will become an industry standard, even surpassing the popularity of ChatGPT by the end of 2024.
Llama 3.1 FAQ
Q: What is Llama 3.1?
A: Llama 3.1 is the latest open source language model developed by Meta, with 405 billion parameters.
A: According to Meta, Llama 3.1 outperforms GPT-4 on several benchmarks and costs less to deploy.
A: Llama 3.1 will be available on WhatsApp, Instagram, Facebook and via Meta Quest VR headsets.
A: Meta believes that open source enables faster and broader improvement of AI models, similar to Linux’s effect on operating systems.
A: Microsoft, Amazon, Google, Nvidia, and several other partner companies.
A: Llama 3.1 has exceptional performance and is open source, allowing developers to tailor it to their specific needs.
A: Meta offers a limited version in free prompts, with the possibility of reverting to a less powerful version once the limit has been exceeded.
A: Currently, Llama 3.1 is not officially available in France, but a VPN can allow access to it from the United States.
A: Version 3.1 will be available via WhatsApp and the Meta AI site in the United States starting this week.
A: Llama 3.1 is the most powerful open source language model to date, with capabilities and performance surpassing even GPT-4.
To read Antigravity 2.0 : Exploration du nouvel eldorado des agents IA autonomes
