Stability AI Releases FreeWilly Language Models | | | | Turtles AI

  While aiming for open access and ethics, the research from #Stability AI's #CarperAI lab has now resulted in two impressive new language models, #FreeWilly 1 and #FreeWilly2. A synthetic dataset contributed to the high though responsible performance of these language models, which excel at complex reasoning while keeping politeness. There are still areas for improvement in red-teaming to ensure safety, but the initial results suggest a promising advancement towards open AI. Highlights:

• Stability AI releases FreeWilly1 and 2 Large Language Models
• Models were trained on synthetic dataset inspired by GPT-4 approach
• Performance across evaluations shows strengths in reasoning and nuance
• More testing needed to ensure safety and for societal impacts