‘Stability AI,’ the company most known for its Stable Diffusion image-generation and image editing service, recently announced two new Large Language Models (LLMs) named FreeWilly1 and FreeWilly2. What is unique about the FreeWilly LLMs when compared to traditional LLMs is that the FreeWilly models are trained using synthetic data and concentrated datasets.
The name for the models, FreeWilly, comes from the story about the baby whale in the ’90s. The relevance of the whale to the LLM is that the FreeWilly LLMs are based on Microsoft’s ‘Orca’ AI training methodology. However, the FreeWilly models only use 600,000 datapoints, or roughly 10% of the Orca method, which means they are essentially baby whales. Stability AI is aiming to show the efficacy of smaller, more focused LLMs, rather than all-encompassing LLMs, both for reducing environmental impact and for ensuring accuracy of results on a smaller scale.
Image Credit: Shutterstock
