Large artificial intelligence (AI) models now seem to be powerful enough to create new small AI tools without any human intervention.
Teams of scientists from several campuses of the Massachusetts Institute of Technology and the University of California, as well as artificial intelligence technology company AIZIP, reportedly wrote in a research report last Friday (December 15) that they can make large AI models largely self-replicating.
Yan Sun, co-founder and CEO of AIZIP, said in an interview, "Right now, we're using big models to build smaller models, and it's like an older brother helping its younger brother improve." ”
This is the first step towards self-evolving AI, and it is also the first step to showing that AI models can build other AI models. ”
Yubei Chen, another co-founder and researcher at AIZIP, added, "Our technology is a breakthrough because for the first time, we have designed a fully automated pipeline that allows a large model to design an AI model on its own without human intervention. ”
Yan Sun on the left and Yubei Chen on the right).
Small AI models.
Large language models like ChatGPT reportedly cost $700,000 per day to run, while small AI models, often referred to as micro-machine Xi or tinyml, can run at a low cost and can be carried around on the device.
TinyML can be used for any specific task, from facial recognition to hearing aid devices and home appliances.
"We're working on very concrete tasks, focusing on very low-cost solutions that can be used in every corner of the world," Sun said. ”
He added, "We're bringing intelligence into everyday life to make life safer." ”
The team of scientists initially focused on creating micro-AI, but they say the design process can now be automated by smarter AI.
"In the future, we believe that large and small companies will work together to build a complete smart ecosystem. ”