Berliner Tageblatt - Meta pushes global AI vision amid race with OpenAI, Anthropic and China

In a long essay published Monday about the future of artificial intelligence, Zuckerberg called for the US to compete against China, prevent "government tyranny" over AI and ensure that "superintelligence" technology will become available to everyone.

Superintelligence is a theoretical point when AI's capabilities exceeds human intelligence.

The solution, he argues, includes "open" weight models, which allow users to download the rules that govern how an AI system behaves. Unlike truly open-source software, however, open weight models do not make all of its underlying code publicly viewable.

Competitors like OpenAI and Anthropic largely focus on closed AI models, which are considered black boxes that users cannot inspect themselves.

Meta has poured billions of dollars into its AI division, known as Meta Superintelligence Labs, as it attempts to keep up with OpenAI and Anthropic -- the makers of ChatGPt and Claude, respectively.

Meta also announced on Monday a new open weight model called Glimmer, which it developed in part from Muse Spark, a closed-source model it announced in April and updated last week.

Glimmer uses less computing power than other models but is still muscular enough to operate AI agents or software that can perform tasks without human intervention, Meta said in a blog post.

Developers in China, including Alibaba, DeepSeek and Moonshot, have created advanced and more affordable open models in recent months, spurring American companies to lower prices and build their own cheaper models.

Zuckerberg also took aim at other labs that are "focused on building AI for companies, governments or other institutions," arguing that this would tip the balance of power in favor of institutions instead of individuals.

Anthropic and OpenAI both have lucrative government contracts.

"Rather than centralizing superintelligence, we should distribute it widely and give every person the ability to direct it," Zuckerberg wrote.

- Rules and lawsuits -

Meta has 3.6 billion users across its apps which include Facebook, Instagram and WhatsApp.

But concern over app usage by American youths has soared. Meta was recently ordered to pay nearly $1 billion in damages in a New Mexico case.

And on Wednesday, jury selection begins in a California court for a trial brought by four US states seeking a staggering $1.4 trillion in damages from Meta.

The company faces potentially thousands of costly lawsuits nationwide about alleged harms to children who use Facebook and Instagram.

Zuckerberg's essay, and Meta's new models, also come amid US debate about how to regulate AI developers, in addition to growing local pushback against building data centers that power the technology.

Facing pressure over the novelty of their capabilities, President Donald Trump in June ordered the setting up of a voluntary review process of advanced AI models before their release.

But his adminstration missed its own August 1 deadline to announce the details.

"Given how dynamic and competitive the AI landscape is, I think it will be more productive to have a close proactive collaboration between frontier labs and the government rather than a rigid process and review timeline" in all cases, Zuckerberg wrote.

A flexible US regulatory environment for open models would give Meta an advantage over OpenAI and Anthropic, and help it compete with Chinese developers.

"Our goal should be for American open source models to be the best globally. This requires removing the hurdles that make it harder for American open source models to compete," according to Zuckerberg.

Even a one-month delay could add "significant risk to American leadership" while letting foreign models race ahead, he added.

With public opposition to data center development growing, Meta will launch a $1 billion fund to directly support communities around its facilities, a spokesperson told AFP.

Meta expects to spend up to $145 billion this year, nearly doubling 2025 spending, as it invests heavily in AI.

L.Janezki--BTB