The grand pact to reach a consensus on how to develop artificial intelligence is losing supporters. The latest to distance himself is Mark Zuckerberg, Meta's CEO, who aims to emerge as one of the winners of the month in the ever-unstable tech world.
The Facebook founder has been clear in a message on X where he explained his stance: big alliances are not needed, what is necessary is for Anthropic and OpenAI to take the security of their models seriously and invest more in that than in recursive self-learning, that is, in achieving an AI that improves itself with the control problems that entails.
"Committing a significant portion of your computing power to serving people instead of rushing to achieve recursive self-learning is one of the best ways to develop this technology safely. Meta has committed to this, and other labs can too," emphasized the entrepreneur.
In his lengthy message, the executive stated that the tech company has delayed for months the launch of Muse, its personal AI assistant that is succeeding in the United States, to ensure that it was secure enough and now has external evaluators of its models, one of the measures proposed by Dario Amodei, CEO of Anthropic.
At the same time, he expressed confidence that the AI market itself would push companies to opt for a safer path when developing AI. "I believe that trust and alignment (of models) are quickly becoming the most important capabilities to differentiate agents and models. Those who do not focus on alignment will be left behind," he predicted.
The executive's words contrast with Meta's willingness to reach sectoral agreements on other issues such as restricting the use of social networks by minors following their pact in the United States, an agreement that had several conditions specifically subject to its adoption by Google and TikTok.
At the same time, the various scandals regarding the company's data handling do not help the vision of being a guardian of the ecosystem and a secure company that Zuckerberg and his team seek to convey, dedicating the bulk of the Muse presentation to selling security.
Google's secret project
The big question now in the ecosystem is who will be the next company to announce a major breakthrough in an environment increasingly filled with speculation, pointing to the labs as another of the millennium's problems beyond Navier-Stokes but they dare not publish it.
Other major speculations point to Google as a potential black swan in the debates that cross-cut AI.
The president of its AI unit, Demis Hassabis, supported Amodei's message and shared his own manifesto calling for a standard for this technology. However, recently, the researcher was moved to a less executive role amid growing tension over resource usage within the company, leading to a talent drain more aligned with Hassabis leaving the company to found a company focused on researching the application of AI to science.
These changes are attributed to the growing gap that has opened up between Gemini, the leading model at the beginning of the year when it seemed that Google had everything to win the AI race, and the various models driven by OpenAI and Anthropic.
Now, the company's new direction is not as clear as when Hassabis was the sole thinker, and Reuters already reported that Sergey Brin, Google's co-founder, has taken on an important role in the AI team, although he does not formally hold a position. Among Brin's main obsessions was precisely achieving being the first lab to achieve the mentioned recursive self-learning, a discovery that would surely reverberate throughout the AI ecosystem.
The executive is famous for his ability to drive projects, sometimes beyond reason as was the case with the failed Google Glass, although another of his bets, Waymo, has allowed the tech company to lead in autonomous driving.