A new research initiative by Sakana AI Labs, an AI research company based in Tokyo, Japan, and founded by former Google researchers, is using large language models (LLMs) to automate AI research and discovery. This innovative approach aims to improve the training of foundation models by leveraging LLMs to generate hypotheses and code. The process has led to the development of DiscoPOP, a novel algorithm that optimizes large language model outputs by merging logistic and exponential methods, surpassing traditional techniques. This advancement is seen as a significant step towards achieving Artificial General Intelligence (AGI), defined as an AI capable of performing human-level tasks.