In a significant leap forward for artificial intelligence, researchers have unveiled a breakthrough that allows large AI models to autonomously create and train smaller AI tools without human intervention. It’s a development that echoes scenes from science fiction, prompting some to draw parallels with the iconic Terminator franchise.
The traditional approach to AI involves using large, powerful models like ChatGPT to perform a myriad of tasks. However, these models cannot respond to real-time sensor data and make customized adjustments. Now, researchers from MIT, University of California campuses, and AI technology company Aizip have unveiled a system that can ingest real-time sensor data and create a slightly different AI replica.
This innovation is akin to the concept of “multiplicity,” where a person clones themselves for specific tasks. UC Davis Professor Yubi Chen, founder of Aizip, envisions a future where small AI models, created through this system, can transform everyday objects like running shoes, headsets, and IoT devices by making precise adjustments based on real-time data.
The process involves building smaller AI models from larger ones, a sort of “big brother helping its smaller brother to improve.” The technology promises a new era where every item can become a smart, evolving companion, responding dynamically to the needs of users.
While the potential applications are exciting, there are concerns about unintended consequences. As the AI community celebrates this technological leap, there’s a need for caution. The self-replicating nature of these AI models, though on a smaller scale, raises questions about the direction they might take and the lessons learned from fictional scenarios like Skynet.
Yan Sun, CEO of Aizip, emphasizes the breakthrough’s potential to embed intelligence into daily life, making it safer. The focus is on creating specific, low-cost solutions for global use, bringing AI into various corners of the world.
In a bid to encourage diversity and inclusivity, the researchers invite applications from founders of all backgrounds, irrespective of their geographical location or experience level. The program, named Converge 2, will run from March 11 to April 19, 2024, with a commitment of 4-6 hours per week, offering a unique opportunity for innovators to explore the possibilities of self-evolving AI.
As we stand at the intersection of technological advancement and the potential pitfalls of unforeseen consequences, the AI community treads carefully into uncharted territory. The collaboration of large and small AI models promises to build a complete intelligence ecosystem, ushering in a new era in the world of artificial intelligence.