ParentFull threadgbasin·you don't need access to the weights though, unless I missed something? this is akin to the various groups using GPT-3.5/4 to generate data for fine-tuning smaller modelsView on HN