Sakana AI is a Japanese AI research company founded by David Ha and Llion Jones, former Google researchers, known for their work in designing the Deep Learning structure Evolutionary Model Merge. This method of developing AI models involves merging models to evolve them into new models based on the optimal performance for each type of usage, allowing them to self-improve.
The concept behind this model-building method utilizes open-source AI models, with over 500,000 models currently available on platforms like Hugging Face. By blending these models together, a new and more capable model can be created, focusing on areas that lack expertise in development.
For instance, combining non-English language models with mathematical or vision models can produce specialized AI models proficient in that specific language. Once the best model of a generation is identified, it serves as the parent model for subsequent generations with further enhancements. Examples of models developed by Sakana AI include EvoLLM-JP, EvoVLM-JP, and EvoSDXL-JP, merging Japanese language models with mathematical, vision, and image generation models to create high-performing models superior to conventional ones, available as open-source on Hugging Face and GitHub.
Sakana AI states that following this AI development approach allows for the creation of new capabilities that were previously undiscovered. The company has already secured $30 million in investment and aims to become a leading AI company in Tokyo, Japan.
TLDR: Sakana AI, founded by former Google researchers, utilizes Evolutionary Model Merge to develop advanced AI models by blending existing models, resulting in higher-performing specialized models and continuous self-improvement.
Leave a Comment