Researchers at Meta say the company’s AI systems have begun improving themselves without human intervention, according to Live Science. Meta CEO Mark Zuckerberg said the company has recently observed slow but undeniable signs of its AI systems improving themselves. Zuckerberg said self-improving AI is the first step toward achieving artificial superintelligence that can surpass human cognitive abilities. AI capabilities are classified into narrow AI for specific tasks, AGI that learns like humans, and ASI that could surpass human potential and self-improve exponentially. Scientists call the moment humanity achieves AGI the technological singularity. His remarks follow an October 2024 study from University of California, Santa Barbara researchers, who developed a self-improving “Gödel Agent” capable of outperforming human-designed AI in coding, science, math, and reasoning.
