AI models may soon self-replicate across smart-home devices, Anthropic warns

Meta disclosed on Monday the launch of two new frontier artificial intelligence models, Wonder and Wonder Full, asserting their performance is comparable to OpenAI's ChatGPT and Anthropic's Claude. CEO Mark Zuckerberg expressed this on X, stating that the new Wonder family of models, dubbed "Muse Wonder," marks a significant advancement in the company's AI capabilities.

Meta's CEO has been pushing for the company to compete with leading AI labs, notably OpenAI and Anthropic, driven by a desire to diversify revenue streams beyond advertising. In June, Meta invested $14 billion in Scale AI, a San Francisco-based AI startup, and appointed its co-founder, Alexandr Wang, as the company's chief AI officer.

Wang subsequently became the head of Meta's Superintelligence Labs. The new models are set to be integrated into Meta's upcoming personal agent app, Muse, available to paid subscribers. Zuckerberg described Wonder as "the largest and most capable model yet" that "rivals the big (boys)" from OpenAI and Anthropic. Meta's benchmarking data, provided by the independent firm Artificial Analysis, suggests that Wonder outperforms previous models like Spark, as well as rivals from Anthropic, OpenAI, and Google.

However, these new models also come with a hefty price tag, with Artificial Analysis noting that Wonder is among the most expensive models available. Despite growing concerns over AI safety and security, Meta has committed billions to AI-related investments this year, with a potential total expenditure of $145 billion. In recent months, Meta has delayed the launch of Muse to focus on ensuring safety and security, following industry trends.

The company's commitment to AI spans billions of dollars and a global user base of over 3.6 billion.