@transitionalaspect Well now I'm trying to imagine a sampling technique that could justify the "snowblow" name. Chopping up things from over here & spraying them on top over there? Maybe that's when you take an initial dataset, use it to train an LLM, and then have the machine regurgitate mashed-up synthetic examples?(For anyone unfamiliar: a snowball sample is when you start with a few documents/persons that meet your criteria & then follow their references/personal connections to others, getting bigger & bigger as you go, like rolling a snowball out of sticky snow.)
