Meta: The AI Comeback Kid (LLama edition)

Meta's fortunes on the rise through its development of open-source LLMs

Happy Friday all,

Meta (previously Facebook) might have been better named 'GenAI', had they known the impact that AI would soon have on technology and humanity.

Their gamble on the 'Metaverse' seemed forced, overlooking the fact that the future is shaped by innovation, not only by a top-down, centrally controlled vision.

Ironically, no one seems to understand this better now than Mark Zuckerberg — master of the ‘Metaverse’ — who's using AI to steer Meta out of a slump.

Meta’s Comeback

Meta hit an absolute nadir last year, suffering a record loss of $230bn in market value in a single day. A historical setback, it came precariously close to losing its status as one of the top 20 biggest US firms.

No surprise, perhaps, that Zuckerberg was conspicuously absent when President Biden convened an ‘AI safety’ meeting with the CEOs of Open AI, Anthropic, Microsoft, and Alphabet in May 2023.

Adding insult to injury, Sam Altman got a rapturous reception when called to testify ahead of the US Senate — a markedly difference experience from Zuckerberg’s own 2018 mauling ahead of US policymakers.

When it came to AI, Meta was *literally* not in the room; and Zuck was not considered to be in the same league as Sam Altman.

AI: The turnaround catalyst for Meta

But AI has played a seminal role in the reversal of Meta’s fortunes over the past few months. While Meta was not GenAI’s breakthrough star, it is catching up - big time.

In February, Zuckerberg announced a new ‘Generative AI Unit’ at the heart of Meta. Now it is making tracks with its first major project — Meta’s own family of large language models series: the so-called Llama series.

“We are no longer behind in building our AI infrastructure”

Mark Zuckerberg, April 2023

Llama: An open-source triumph

When the inaugural version of Llama was rolled out in February, it was big AI news, mainly because it was introduced open source. While industry giants like Google and Open AI have been developing closed-source LLMs, Meta chose less control for more potential impact.

There’s no doubt that closed-source LLMs offer impressive performance, but they come with certain limitations. One significant disadvantage of closed LLMs is the limited control they provide to users. Given that the underlying architecture and weights of these models are not publicly available, it is difficult to customise and fine-tune them.

But as we transition beyond the ‘incubation’ phase of Generative AI into the era of mass adoption— customisation and fine-tuning of Foundational models is exactly where the enterprise value (and innovation) RE: Generative AI is going to come from.

So when Llama 2 — the improved version of LLama — was unveiled in July in partnership with Microsoft (seriously, Microsoft is everywhere!) it was released open source, with model weights accessible AND free for commercial use.

Already, Meta claims that Llama 2 surpasses other large language models in many external benchmarks, including reasoning, coding proficiency, and knowledge tests. By offering startups and businesses a cost-effective alternative to expensive proprietary closed models, Llama 2 is poised to democratize the AI landscape.

In the best case, LLama 2 will empowering researchers and commercial users to explore and push the boundaries of what AI can achieve — with Meta at the heart of this innovation.

Of course, no one can really move in this space without tapping into the powers that be — the ‘AI Oligopoly,’ I wrote about last week — so Meta has also partnered up with Microsoft and Amazon to make the model available on the IT giants’ Azure and AWS cloud platforms, respectively.

And Zuck’s moves appear to be paying off. Meta is certainly at the top table again. By July, it was included at a second White House AI meeting on AI, committing with six other companies — including Microsoft, Google and Open AI - to develop AI ‘safely.’

Meta’s masterstroke: leading OS LLM development

So perhaps that is the real value of Zuckerberg’s ill-spun foray into the ‘Metaverse’…. It appears he’s learned the biggest lesson of all. When it comes to real change and innovation — it cannot be be centrally controlled or prescribed top-down.

Meta is not alone in this realization. Just yesterday, the Chinese tech monolith Alibaba announced its plans to roll out open-sourced AI LLMs to compete with Meta's Llama 2. This follows the impressive performance from the Falcon LLMs family developed open source by the Abu Dhabi Technology Innovation Institute.

As I wrote at the start of the year, the evolution of open-source AI models will continue to be one of the most important trends for this space.

As the dynamics of the ‘AI model market’ pivot from ‘bigger is better’ to ‘cheaper is better;’ ‘more efficient is better;’ and ‘customizable is better;’ it appears that Meta has pulled a strategic coup by positioning itself as the leader in the development of open-source LLMs.

Watch this space.

For now,

Namaste,

Nina