Meta's Llama enters the generative AI fray with ChatGPT, Bard

Meta, the last of the tech giants to deliver a generative AI system, plunged into the surging market with its own large language model, Llama.

The Facebook parent company released its LLM on Friday in a limited-access program to select participants and with a model card outlining its own development methodology. Meta positioned Llama as a "responsible" alternative to the advanced AI chatbots from the Microsoft-OpenAI alliance and Google's Bard.

Last week, AWS jumped into the fray with a partnership with generative AI vendor Hugging Face aimed at LLM developers. At the time, Forrester Research analyst William McKeon-White noted that the tech giants are compelled to keep up with each other whenever they discover a new market.

"Where one goes, the others have to go. Otherwise they risk losing out on the massive market," he told TechTarget Editorial.

For Meta, which has pioneered open source AI research and development over the years, introducing Llama now ensures that it will not fall behind in the supercharged generative AI market.

"The pace of generative AI will continue to accelerate, and Meta entering with its competitive offering shouldn't come as a surprise," said Daniel Newman, an analyst at Futurum Research. "With its deep roots in AI development and its massive data set, it is set up as a prime candidate to play a role in generative AI's revolution."

ChatGPT captured the world's imagination since it was introduced in November, several years after its more stolid precursor, GPT-3, was released in 2020.

Microsoft -- an early backer of ChatGPT's creator, AI upstart OpenAI -- bet heavily on the technology, investing $10 billion in ChatGPT. The company added some of ChatGPT's capabilities to its Bing search engine and promised to more widely deploy it in the rest of its enterprise software universe.

Meanwhile, both ChatGPT and Google Bard floundered in their early months and weeks, making errors and showing that much work is still needed to make the powerful new tools -- which can write code, create essays and art and conduct intensive research -- safe for wide use.

It is in this context that Meta, in its Llama introduction blog, marketed its own LLM as part of Meta's "commitment to open science" and as a smaller-scale, more useable foundation model that is more democratic than others because it requires less computing power.

Meta said it introduced Llama under a noncommercial license "to maintain integrity and prevent misuse."

Where one goes, the others have to go. Otherwise they risk losing out on the massive market.

With the instability of some of the biggest LLMs, that emphasis on caution and smaller scale is welcome, said Kashyap Kompella, an analyst at RPA2AI Research.

"The recent narrative is that Meta and Google have ceded the AI leadership space to Open AI, as the virality of ChatGPT shows," he said. "But in reality, the AI research at Meta, Google and others has led us to ChatGPT.

"Meta and Google already use several LLMs under the hood," he continued. "LLMs are quite powerful and are to be handled with care. In a sense, I am happy that Meta is not moving fast and breaking things on this."

It remains to be seen which generative AI system will dominate in the fast-expanding market. Information provided through YouChat, a generative AI tool, said ChatGPT is a more open model than Meta's Llama in that it allows developers to customize their experience and adjust the level of complexity of their output. Additionally, ChatGPT lets users moderate the conversation, such as the ability to manually review and flag inappropriate responses.

By comparison, Llama is a more closed model, meaning that the model is pretrained and the user has less control over the output. It is also more focused on providing accurate and factual information, as it relies heavily on search results to provide answers to questions.

While all four of the biggest tech giants are now in the generative AI game, the contest is still in its early stages, Newman said.

"It's a bit of a proverbial arms race among big tech to position their generative AI offerings," he said. "However, we are still in very early days, and I expect a lot more flaws and discovery to come before we see its full potential. But it is the beginning of something really special."