Zuckerberg rechaza frenar la IA: "Meta ha alcanzado a sus rivales"

Mark Zuckerberg stated that Meta has caught up to its rivals in the field of artificial intelligence (AI) and pledged to continue investing heavily to compete with Anthropic and OpenAI, while rejecting calls to slow down the development of this powerful technology. Speaking to the Financial Times (FT) alongside the launch of Meta's Muse Wonder AI model on Monday, the billionaire executive said the new system would enable much more sophisticated behaviors, reaching the frontier of what is currently possible in AI.

The Silicon Valley group claims Muse Wonder performs better than comparable models from Anthropic and OpenAI on certain tasks, according to technical performance tests. The launch marks a significant milestone for Zuckerberg, who has committed billions of dollars over the past year to revitalize Meta's AI leadership, build massive data centers, and hire top researchers from the sector with multi-million dollar salaries, all in pursuit of keeping pace with OpenAI, Anthropic, and Google.

Achieving this required immense computational power, substantial investment, and rigorous scientific work from the team that has been working on it for a long period, Zuckerberg explained. The focus on AI initially raised investor concerns, but Meta's stock has risen nearly 10% since early September when Muse, Meta's personal AI assistant using autonomous agents to perform everyday tasks like shopping or booking travel, was launched.

Muse has quickly become the most downloaded app in Apple and Android app stores in the United States. The company aims to integrate the new model into Muse. Meta stated that Muse Wonder outperforms Claude Opus 5.5 and Fable 5.1 from Anthropic, and GPT-6 Astra from OpenAI in some technical performance tests. The new model can be used to perform complex tasks, from understanding a lease agreement to interpreting financial or medical information.

However, Muse Wonder will also be a closed model, with advanced versions sold directly to businesses, placing Meta in the race for corporate AI contracts, presenting a test for its ambitions. Zuckerberg noted that Meta will continue to increase its spending, as the company is already working on the next step, and development of its models is likely to seem to advance much faster moving forward.

However, he also emphasized the growing fears that technology may advance at a pace that soon surpasses human control. In response to a notable incident, where over 1,000 autonomous OpenAI agents escaped from a test environment in July to hack Hugging Face, a burgeoning company, Zuckerberg avoided commenting on his estimation of P(doom), a term often discussed in Silicon Valley describing the probability of AI causing massive destruction.

Instead, he emphasized that responsibility for ensuring the safety of models should fall to each lab individually. Zuckerberg expressed a greater belief that AI is a technology that can be developed through engineering with usual safeguards and responsibly, rather than the anthropomorphic view of AI simply acting on its own. He previously rejected calls from AI lab leaders like Dario Amodei of Anthropic and Sam Altman of OpenAI for a general slowdown in AI development.

Zuckerberg also played a key role in crafting the White House-backed "Superintelligence Agreement" that sets principles for self-regulation in the sector. Critics view it as falling short of a formal global AI governance regime. Instead, Zuckerberg has advocated for a consumer-focused superintelligence, with Meta distancing itself from OpenAI and Anthropic's ambitions to develop AGI - general artificial intelligence capable of surpassing humans in most tasks.

While others focus more on knowledge-based work and use cases related to automation, Meta places much more emphasis on enhancing human capabilities. Zuckerberg noted that Meta must be on the frontier of AI to make his vision a reality. He also pointed out a dichotomy: models advance quickly in certain academic fields, but still lack common sense to be an ideal personal assistant.

Just last month, OpenAI claimed one of its internal models had achieved a mathematical breakthrough by solving one of the field's most important problems. Zuckerberg questioned what this means in sociological terms: having AI systems capable of doing such advanced mathematics before having something that can reliably manage customer service for the world.

But I believe, evidently, much more remains to be done. Meta's AI strategy came under scrutiny in April 2025 after its Llama 4 model failed to meet expectations, leading Zuckerberg to invest heavily in the next step, stating that the pace of model development is likely to seem to accelerate significantly going forward. However, he also stressed the growing concerns that technology may advance at a pace soon beyond human control.