Sekėjai

Ieškoti šiame dienoraštyje

2024 m. gegužės 20 d., pirmadienis

Facebook Parent's Plan to Win in AI: Free Tech


"Mark Zuckerberg has an unusual plan for winning the artificial-intelligence race: giving away his company's technology free.

Like many of its rivals, Zuckerberg's Meta Platforms is spending tens of billions of dollars on high-end computer chips, top-flight computer scientists and gigawatts of electricity to build the most powerful AI tools it can. Unlike any of those rivals, some of whom made AI announcements last week, Zuckerberg is giving away the fruit of that investment -- Meta's most advanced chatbots and the technology that drives them.

Behind the contrarian strategy is a bet that making Silicon Valley's hottest new technology free will drive down competitors' prices and spread Meta's version of AI more broadly, giving Zuckerberg more control over the way people interact with machines in the future.

"What you don't want in a situation like this is for other people to own the unique technology you don't have access to," said Jerome Pesenti, Meta's former vice president of AI and now the founder of Sizzle AI.

Meta felt the cost of not controlling its own destiny when Apple decided in 2021 to cut off its ability to gather data on their users without asking for permission -- something Meta had relied on to target ads. The company said it took a hit of more than $10 billion in revenue in 2022 as a direct result. The company's stock swooned 26%.

For the AI-giveaway strategy to work, Meta must get its billions of users to look to those free AI services in the same way they flocked to Facebook, Instagram and WhatsApp. It wagers that advertising can come later, as it did in the past. Meta's ability to turn eyeballs into ad dollars is well established, although early user responses to its AI services have been mixed.

The effort to draw users to a free AI product grew more competitive last week after OpenAI launched GPT-4o, a conversational-assistant version of its blockbuster AI model that it intends to give away free. Google launched new versions of free AI models it calls Gemma. But neither is releasing the underlying code of their most powerful tools, to which they sell access to make money. 

OpenAI has raised at least $13 billion, largely from Microsoft. Google said it is boosting capital expenditures by 50% or more this year.

Meta said last quarter that AI could lead it to spend an additional $10 billion in 2024. So far there isn't an obvious way to recoup its spending in the near term. Meta makes its money selling ads on its social-media services.

Meta doesn't sell cloud access to its servers -- one of the main ways its AI competitors plan to monetize the tools called large language models, or LLMs. 

Generative AI tools are able to help write code and provide humanlike answers to queries, among other things.

Its shares plunged 10% last month after Meta detailed its new spending on AI. 

But Meta is seeing early support from some investors and analysts for its argument that it will benefit eventually by releasing its generative AI as "open source." Open-source software is publicly available, meaning Meta is releasing the source code and doesn't intend to charge most businesses to use it.

Meta last month released its most recent generative AI tool -- dubbed Llama 3 -- free for any company to use and repurpose as long as they have fewer than 700 million users.

 And it integrated chatbots based on Llama 3 into Instagram, WhatsApp, Facebook, Messenger and on the web.

"All of our properties are free, we help people connect with each other, and we want to help connect people with AIs that can help them get things done," Ahmad Al-Dahle, Meta's vice president of generative AI, told The Wall Street Journal. "That's always been our playbook. That's what the company ethos is about."

In March, Zuckerberg decided to double down on offering free AI products to users, giving the teams working on the company's chatbot, Meta AI, just one month to turn his vision into reality, according to people familiar with the matter.

Zuckerberg contributed directly to the effort to integrate Meta AI into the search functions of Facebook, Instagram and WhatsApp, making a number of key design decisions. It was a familiar pattern for the chief executive, who has been known to take a hands-on approach on certain projects.

Some competitors and government officials are concerned that by making some of the most advanced AI tools freely available, Meta could help adversaries and hackers. At a summit last fall in the U.K., some officials raised the idea of putting limits on open-source AI models.

Meta says it has been engaged in discussions with officials in the U.S. and elsewhere on AI safety, as well as on national security. The company includes limits on how its Llama models can be used with its open-source license and has added technical guardrails that limit how the model will respond to certain queries.

One advantage of releasing its Llama models free is that Meta can watch a whole community of open-source developers refining the tool -- and then integrate those improvements. Another is that it is a recruitment tool for top AI researchers, many of whom come from academia and prize the ability to keep publishing their work.

Within days of the launch of Llama 3 in April, the open-source community improved the model so that it could take in 20 times as many tokens at a time than when it launched. A token is a unit of text, like a word or a punctuation mark.

If Meta succeeds, it would be able to define its own destiny in a way it was never able to do in the mobile-phone revolution. On mobile, Meta is beholden to two companies -- Apple and Alphabet's Google -- that control the mobile-phone operating systems.

"It seems to be driven by a desire to commoditize the market," said Bryan Offutt, a partner at Index Ventures, who invests in enterprise tech. "They don't want a world where the entire tech community is building the foundations of the next 20 years of applications on OpenAI and Microsoft."" [1]

1.  Facebook Parent's Plan to Win in AI: Free Tech. Rodriguez, Salvador; Schechner, Sam.  Wall Street Journal, Eastern edition; New York, N.Y.. 20 May 2024: B.1.

Komentarų nėra: