Hexo Labs, the research lab developing superintelligence, has launched an open-source agent called SIA, short for Self-Improving-AI.
The company describes SIA as the world’s first agent that teaches itself how to perform tasks and improve autonomously, accelerating the path to superintelligence by 350X according to a benchmark designed by OpenAI.
SIA is designed to operate in continuous loops of learning and adaptation. It generates hypotheses, runs experiments, evaluates outcomes, updates its approach and repeats. Similar approaches have been demonstrated by systems such as Google DeepMind’s AlphaGo, which mastered the ancient Chinese board game Go, and IBM’s Deep Blue, which defeated chess grandmaster Garry Kasparov.
“Today’s AI systems are powerful but share a fundamental limitation: every meaningful leap still depends on intervention of human experts to decide what to try next, interpret results, and refine direction,” said Kunal Bhatia, CEO and Co-Founder, Hexo Labs.
“But superintelligence will not emerge from static models. SIA learns from itself through execution and compounds its capability with every cycle.”
The Hexo Labs research team plans to apply SIA to complex problems across science, business, engineering and other domains that require continuous iteration to improve outcomes. The company works with leading research scientists at Stanford University, the University of Oxford and the University of California Santa Barbara.
“Hexo Labs is on the cutting-edge of AI research,” said Eric Wang, Asst. Prof. at the University of California Santa Barbara. “Hexo Labs’ work towards Recursive Self-Improving AI could be gamechanging.”
