 "Meta's days of giving free AI may be ending as monetisation plans loom")
By Parmy Olson
Like any good investor Mark Zuckerberg likes to make a return, but he’s willing to play the long game. It took a decade from spending $19 billion on WhatsApp before he finally put ads on the messaging app. Now, as he invests $65 billion this year on artificial-intelligence infrastructure and stokes a talent war to hire the field’s brightest minds, expect him to monetize that endeavor too by pivoting away from the free AI project that has left Wall Street scratching its head.
Llama is the flagship AI model Meta Platforms Inc. built to catch up with ChatGPT. Unlike those made by Alphabet Inc.’s Google and OpenAI, it’s billed as “open source,” meaning its blueprints are plastered on the internet for anyone — including his competitors — to copy and build upon. Why? Zuckerberg has said it’s to make AI accessible, so that “everyone in the world benefits” — a line I didn’t buy that from the guy who built his empire on harvesting user data to sell to advertisers.
What’s more plausible is that Llama is a launching pad for something more commercial, and has now served its purpose. Open sourcing the model gave Meta a free research and development workforce, as thousands of developers bought into the ideal of democratizing AI for all and helped improve Llama for free. It’s also made Meta a magnet for talented researchers who might otherwise go to competitors, and who want to see their contributions reach millions of users immediately rather than get locked behind Meta’s products. And it’s burnished Meta’s reputation as being on the side of the angels in the AI race, even when Llama’s terms of use were restrictive and didn’t fit the technical definition of open source.
Critics who’ve accused Meta of “open-source washing” were right to believe this wasn’t going to last. Zuckerberg softened his stance on the issue already last year, telling the Dwarkesh Patel podcast that it was wrong to be “dogmatic” about open-source software and that if it became irresponsible to give his AI away in the future, “then we won’t.”
There’s no clear answer on whether open-source AI is good or bad for society. Sure, it addresses a growing concentration of power among Silicon Valley companies; but as free models become more capable, giving them away raises the risks of misuse. Chinese military researchers have used Llama to build their own intelligence tool called ChatBIT, for instance, using it to gather data and make operational decisions. Researchers also created a “BadLlama” model with no safety features, something they wouldn’t have been able to do with closed systems like ChatGPT.
To the extent that Zuckerberg — currently the world’s second-richest person after Elon Musk, worth $256.4 billion according to data compiled by Bloomberg — takes occasional advice from Washington DC and policy makers overseas on AI, he will almost certainly have been told to think carefully about making general-purpose AI — software that’s smarter than humans — free for anyone in the world to use.
If and when he does, he won’t be the first to pivot away from a free-and-open strategy. OpenAI started life as a non-profit that pledged to share its research with all, until it started taking large sums of investment from Microsoft Corp. and became more secretive. It's also common practice for tech companies to give products away before monetizing them once they've blitzscaled their way to market dominance.
Wall Street wants to see clearer returns on Meta’s enormous AI spending, and the clues that Zuckerberg will comply are there. Just as OpenAI’s hiring of Fidji Simo signaled it was preparing to launch an advertising business, Zuckerberg’s hiring of Scale Chief Executive Officer Alexandr Wang suggests he wants to monetize the ecosystem he’s built.
Zuckerberg has also been telling some of the star researchers he’s trying to recruit from companies like OpenAI and Google that he won’t open source forthcoming AI models as they reach “superintelligence,” a threshold whose definition is still unclear but refers to AI that surpasses human intelligence in most tasks. Wang has said much to same to the researchers he’s wooing over dinners and meetings.
Don’t be surprised when Zuckerberg announces some time over the next year that Meta’s newest AI models are too powerful to be given away. His open-source AI experiment will have served its purpose: attracting talent, free development and positioning Zuckerberg as AI's reluctant hero. But with billions invested and Wall Street demanding returns, the free ride will end.
Disclaimer: This is a Bloomberg Opinion piece, and these are the personal opinions of the writer. They do not reflect the views of www.business-standard.com or the Business Standard newspaper