Meta has claimed that its AI systems are beginning to show signs of self-improvement, a concept known as recursive self-improvement. This process allows AI to enhance its capabilities without human intervention. Evidence of such technology exists, exemplified by Nvidia's Voyager, a Minecraft bot capable of rewriting its own code, and Google DeepMind's AlphaEvolve, an AI system seen as a significant step toward artificial general intelligence. However, Zuckerberg's assertions lack specific details and substantial evidence to support the claim of breakthroughs in self-improvement.
In recent months, Meta has begun to see glimpses of our AI systems improving themselves, which Zuckerberg suggests is a precursor to developing superintelligence.
Recursive self-improvement involves AI systems developing the ability to improve themselves without human intervention, and has been observed in limited domains like the Minecraft bot Voyager.
Nvidia and a consortium of American universities built Voyager, which continually learned to rewrite its own code using OpenAI's GPT-4, showcasing self-improvement.
Google DeepMind's AlphaEvolve is another self-improving AI system believed to be a significant step toward artificial general intelligence or superintelligence.
Collection
[
|
...
]