Will Llama 3 Enable Meta AI to Compete with OpenAI & Google in 2024?

Why Trust Techopedia

Last week, Meta made some big announcements that will strengthen its AI ecosystem. The organization unveiled its next-generation Meta Training and Inference Accelerator (MTIA) chips.

These chips perform 3x better than the previous iteration and will enable users to run more sophisticated models. It will also reduce reliance on Nvidia GPUs.

Meta also confirmed that it will be releasing two versions of the open-source Llama 3 large language model (LLM) before a larger version is released in the summer, an employee told The Information.

This model is expected to have around 140 billion parameters and will power the virtual assistant in Meta AI glasses – Ray Ban smart glasses.

Meta’s president of global affairs at Meta AI Day London confirmed some of this information.

“Within the next month, actually less, hopefully in a very short period of time, we hope to start rolling out our new suite of next-generation foundation models, Llama 3,” Clegg said. “There will be a number of different models with different capabilities, different versatilities [released] during the course of this year, starting really very soon.”

These developments come less than a month after X.AI and Elon Musk announced the release of the Grok 1.5 LLM, which approached GPT-4 level performance.

Advertisements

In this article, Techopedia looks at Meta AI’s role in the generative AI market to see where it stands ahead of the release of Llama 3.

Key Takeaways

  • Meta released the next generation of the Meta Training and Inference Accelerator (MTIA), the next in its family of custom-made silicon chips designed for Meta’s AI workloads.
  • The company confirmed that it will be releasing two versions of the open-source Llama 3 large language model (LLM) before a larger version is released in the summer.
  • Lightweight versions of Llama 3 have the potential to be more accessible than larger models like GPT-4, making them cheaper to train and run.
  • Llama 3 looks promising, but it’s unlikely to overtake GPT-4 or Gemini anytime soon.

Can Meta Go Head to Head with The Top Dogs?

With the launch of Llama 3, Meta is preparing to go head-to-head with the top dogs in the generative AI market, such as Anthropic, Google, and OpenAI.

“Our goal over time is to make a Llama-powered Meta AI be the most useful assistant in the world,” said Joelle Pineau, Vice President of AI Research at Meta AI Day London. “There’s quite a bit of work remaining to get there.”

Llama 3 will be fighting an uphill battle, not just because of the capabilities of Gemini and GPT-4 (or GPT-5 later down the line), but because the generative AI market is so competitive.

However, some commentators believe that Meta is well-placed to snatch the lead.

Tom Allen, CEO and founder of The AI Journal, said in an email to Techopedia:

“With their announcement of Llama 3 coming this summer, I’d put them ahead of the pack.

 

“They have a massive opportunity ahead of them with the Meta Rayban glasses that are getting rolled-out updates. Plus, the glasses have got a leap up as people are already familiar and comfortable with the idea of wearing Ray Ban glasses. The assistant in the glasses is going to be powered by Llama 3 with the two going hand in hand and [will] improve another line of Meta’s revenue streams.”

Allen also highlights that Meta AI has access to the right data across social media sites like Facebook and Instagram, providing a goldmine of data for training its systems.

On the other hand, if we consider that the first two Llama 3 models being released won’t be multimodal, it is unlikely that they will displace popular models like Gemini and GPT-4, at least in the short term, as they offer support for support text, image, and voice inputs.

That being said, these lightweight versions of Llama 3 have the potential to be more accessible than larger models like GPT-4, making them cheaper to train and run via on-device applications and small devices like smartphones.

A Brief History of the Meta AI’s Llama

When Meta released Llama in four different sizes (7B, 13B, 33B, and 65B parameters) in February 2023, the model received significant attention because it was open-source.

At the time, a large company like Meta invested in developing a powerful open-source model rather than a black-box proprietary model, which provided a much-needed alternative to the closed development approach of providers like OpenAI.

The smaller versions of the model also meant that developers could train and run their own models without spending a fortune on more powerful compute infrastructure. Each version could also be finetuned to improve its performance.

Although Llama didn’t reach the hype of ChatGPT, it still gathered plenty of interest, with Meta receiving over 100,000 requests for access to the model.

However, it was with the release of Llama 2 in July 2023 that the gap between open-source and proprietary AI started to close.

The model, available with 7 billion, 13 billion, and 70 billion parameters, outperformed open-source models on most benchmarks tested and attracted enough attention that by September 2023, over 3,500 enterprise projects were based on Llama 2 models.

Participants embracing Llama 2
Participants embracing Llama 2. Source: Meta AI

Where Does Meta Fit into the LLM Market?

While Meta’s Llama 2 doesn’t outperform GPT-4 or offer multimodal capabilities, it has a distinct advantage: it’s more lightweight and cheaper to run.

By some estimates, Llama 2’s cost per paragraph summary is 30x less than GPT-4 while maintaining equivalent accuracy.

That being said, the open-source nature of Llama 2 may also be holding it back.

Unlike GPT-4 or Gemini, which you can access online via a web app, users need to download Llama 2.

Although this isn’t prohibitively difficult, it creates a less seamless user experience than entering Gemini or ChatGPT into Google or your web browser to load up ChatGPT or Gemini.

Likewise, the lack of multimodal capabilities makes Gemini and ChatGPT better choices for users who want to interact with a virtual assistant that can process text, voice, audio, and image input.

If we turn our attention to the open-source LLM market, we can see that although Llama 2 has been a high performer, it has been outperformed in certain areas by competitors like Mistral AI. For instance, Mistral 7B outperforms Llama 2 13B on all benchmarks and Llama 1 34B on many benchmarks.

On the other hand, the release of Llama 3 and the connection with Ray Ban smart glasses are major wild cards.

If executed correctly, Meta generative AI capabilities will enable users to blend the physical and digital world as part of a new multimodal LLM experience, and Meta’s social media empire provides access to all the data necessary to do just that.

The Bottom Line

Meta remains a key player in the LLM market, but its efforts so far have focused predominantly on enriching the open-source model ecosystem.

Llama 3 looks promising, but it’s unlikely to overtake GPT-4 or Gemini anytime soon.

FAQs

What is Meta AI?

What is Llama 2?

When is Llama 3 going to be released?

What is the latest AI model of Meta?

What is Meta doing in AI?

Can Meta AI generate images?

Advertisements

Related Reading

Related Terms

Advertisements
Tim Keary
Technology Specialist
Tim Keary
Technology Specialist

Tim Keary is a freelance technology writer and reporter covering AI, cybersecurity, and enterprise technology. Before joining Techopedia full-time in 2023, his work appeared on VentureBeat, Forbes Advisor, and other notable technology platforms, where he covered the latest trends and innovations in technology. He holds a Master’s degree in History from the University of Kent, where he learned of the value of breaking complex topics down into simple concepts. Outside of writing and conducting interviews, Tim produces music and trains in Mixed Martial Arts (MMA).