Meta unveils first AI model from costly superintelligence team

Meta CEO Mark Zuckerberg acknowledged record-low morale within the company's Applied AI team in a Friday memo, saying the company has made mistakes and will likely make more, according to a Wired report. The team, created in March to support the Superintelligence Labs, comprises 6,500 staffers. Three anonymous employees told Wired that weekly tasks such as generating puzzles to test AI model reliability are "soul-crushing," with one comparing the environment to a gulag. An employee-only presentation was reportedly interrupted by a worker criticizing a Meta AI executive. Meta's AI-focused restructuring has included thousands of layoffs, increasing workloads for remaining staff. Over 1,600 employees signed a petition opposing new monitoring software that tracks keystrokes and clicks to train AI. Chief Technology Officer Andrew Bosworth wrote in an April memo obtained by Reuters that the vision is for agents to primarily do the work while humans direct and review them. Zuckerberg promised last month that mass firings would cease for at least seven months and called the grunt work critical to advancing models. The company is also facing scrutiny after Wired reported Meta moved to infuse facial recognition technology into its smart glasses.