In the eyes of many, the birth of ChatGPT marks a great transformation in the field of AI.

 

At the same time, this globally popular AI chatbot has also opened the curtain on the competition in the large model field among numerous AI companies around the world. Among them, the large technology company Meta is determined to win this race.

 

It is understood that in recent years, the company has been working hard to introduce generative AI products to billions of users to challenge the leading position of OpenAI, the creator of ChatGPT, in related technologies, including transforming computing infrastructure, integrating research and product teams, and other aspects.

 

In addition, the virtual assistant Meta AI launched in September 2023 has also been built into the search boxes of software such as Instagram, Facebook, WhatsApp, Messenger, etc., facilitating users to easily obtain answers to targeted questions using this AI tool.

 

However, if Meta wants its AI assistant to truly become a competitor to ChatGPT, it must have an underlying model that is as good as or even better in performance than OpenAI. In this regard, Meta has built the Llama series of models.It is reported that on April 18, 2024, local time, the company launched its latest product, Llama 3. According to the introduction, this version is divided into the following two models: one is Llama 3 8B with 8 billion parameters, and the other is Llama 3 70B with 70 billion parameters.

From the perspective of model performance, Llama 3 not only surpasses similar models such as Mistral 7B and Gemma 7B developed by other companies in several core benchmark tests that are currently popular, but also performs exceptionally well in tasks such as reasoning, generating code, and instructions.

"Meta has equipped Llama 3 with new computer coding capabilities and provided it with images and text during training," said Meta's Chief Product Officer Chris Cox to the media.

However, it is worth noting that currently, this model can only output text, and subsequent versions will provide more advanced coding and reasoning functions, such as having a longer context window.

According to a blog post published by Meta, the company also plans to release a version with multilingual and multimodal capabilities in the near future, which means that Llama 3 is expected to generate both text and images.Cox told the media: "The ultimate goal of Llama 3 is to help users get out of trouble and make their lives easier, whether it's interacting with businesses, writing, or planning trips."

In fact, although Meta has now provided a download method for Llama 3, which is free for developers to use, the company prefers to use the term "open" rather than "open source" to define the model.

The reason is that even though Meta allows developers to use Llama 3 for academic research or commercial development, it is not as unconditional as users might think.

In other words, Meta prohibits developers from using the model to train other generative models, and requires companies with "more than 700 million monthly active users" to apply for special permission from it, and Meta will decide whether to grant this permission.

At the same time, according to Meta's Chief AI Scientist Yann LeCun, Meta is training a Llama 3 model with more than 40 billion parameters, which may surpass the current best closed AI models in the world, including OpenAI's GPT-4 and Google's Gemma.Currently, the aforementioned two Llama 3 models have been integrated into Meta AI, and users can access them through the standalone website "Meta.ai".

In this regard, Meta's CEO Mark Zuckerberg posted on Instagram, saying: "With this new model, we believe Meta AI is now the smartest AI assistant you can use freely."

Specifically, Meta AI can help users complete tasks such as recommending restaurants, planning trips, and polishing emails. Even if you use Meta AI's "Imagine" function, you can generate images with very high clarity more quickly.

It is introduced that the image appears at the moment the user types, and it will change "every few letters entered".

It is worth mentioning that Meta AI has only been launched in the United States so far, but it is being launched in English versions in countries such as Australia, Canada, and Singapore.In my view, when many people think of the main AI assistants used today, they don't think of Meta AI, Zuckerberg said to the media. "But I believe it is time for us to start introducing it to many people, and I expect it to become a rather significant product."

In fact, although Meta has regained the top spot on the open-source leaderboard with Llama 3, in the fiercely competitive AI model landscape, this is not the final outcome.

After all, it has been publicly reported that OpenAI is preparing to develop GPT-5, which may help the company surpass other competitors once again. And according to Zuckerberg, Meta is also planning to put the development of Llama 4 and Llama 5 on the agenda.

As he said to the media: "Our goal is not to compete with open-source models, but to compete with all AI in the world and become the world's leading AI."