Meta developing an AI model more powerful than LLaMa 2

A Meta team, hand-picked by Zuckerberg, is working on the new AI tool.
Sejal Sharma
Meta icon on black
Meta developing new AI tool

Dima Solomin /Unsplash  

Meta will unveil a superior artificial intelligence model in 2024, which is touted to be on par with the most powerful model created by OpenAI, the company that birthed ChatGPT and is backed by Microsoft, reported The Wall Street Journal.

WSJ spoke to people familiar with the matter, most likely Meta insiders, who said that the new model would be two times more advanced than Llama 2, the open-source large language model launched by Meta in July and distributed by Microsoft's cloud Azure services.

Interesting Engineering had reported that developers would use LLaMa 2 to design and develop AI tools and would be in direct competition with the likes of Google’s Bard and OpenAI’s ChatGPT. LLaMa 2 has been trained on 70 billion parameters, 40% more data in comparison to LLaMa 1, which Meta launched in February.

While OpenAI hasn’t revealed the size of the data its models GPT-3.5 and GPT-4 have been trained on, it is estimated to be around 1.5 trillion parameters.

Staying true to its open-sourced philosophy 

Just like LLaMa 2, the new system would also be used to help other companies and developers build services that produce sophisticated text, analysis, and other output, as mentioned in the WSJ report. There is no point in guessing, but it is rumored that just like all other AI models that Meta releases, the new model would also be open-sourced.

The group which is building the model has been hand-picked by Meta CEO Mark Zuckerberg. The team’s focus is on developing generative AI tools that can produce human-like expressions. Meta will begin training the new AI system by early 2024.

Meta believes that the new model will be as powerful as GPT-4, OpenAI's most impressive LLM yet.

Maintaining a competitive edge

Meta is competing with the likes of Google, OpenAI, Microsoft, and Apple, all in a deep race and a rush to invest their resources and finances in developing the latest artificial intelligence models. 

To build such sophisticated tools, companies require extremely powerful chips for AI training. Meta is currently building up the data centers and acquiring the world’s most powerful chip out there –NVIDIA H100s – to train their models.

SemiAnalysis, a semiconductor research company, reported a few days ago that Google’s upcoming generative AI model, dubbed Gemini, would be five times more powerful than the most advanced GPT-4 models on the market.

Interesting Engineering reported earlier that Apple is also in the fight. The company is reportedly spending millions of dollars a day to build artificial intelligence tools. Their teams are working on introducing prompts for Siri, Apple’s voice command digital assistant. With a prompt, Siri would be able to produce GIFs using the last five photos in a user’s phone.

Apple is also building an LLM framework dubbed 'Ajax,' and already has an in-house chatbot called 'Apple GPT', which is being used and tested by its staff.

Add Interesting Engineering to your Google News feed.
Add Interesting Engineering to your Google News feed.
message circleSHOW COMMENT (1)chevron
Job Board