Researchers from Sakana AI have developed a groundbreaking methodology that uses evolutionary algorithms to merge models from HuggingFace, enhancing the capabilities of large language models (LLMs) such as understanding Japanese. This innovative approach, termed 'evolutionary model merge,' is considered a form of sophisticated model surgery and requires significantly less computational power than traditional LLM training methods. Additionally, researchers at Stanford proposed FrugalGPT, a cost-saving method that sequentially calls pretrained LLMs from least to most expensive, stopping when a satisfactory answer is provided. The research has sparked interest and discussions within the machine learning community, with some viewing it as a novel way to advance the field beyond the limitations of pretraining work.