Do folks believe in this? Is it possible now? We will prob need new techniques for scalable validation to synth data doesn't bake in self-reinforcingly worse bias and inaccuracy.

not sure tbh. this is a good read https://www.interconnects.ai/p/llm-synthetic-data describing the stage we're at now where the synthetic data generation isn't becoming a recursive phenomenon where synth data gets to grow exponentially relative to organic data

Wanna be the very best. Building @moxiescout. ゼニゼニ

Can you think of a world where this doesn't happen?

It is and self-driving car companies are already doing it to improve their software. See:

It is and self-driving car companies are already doing it to improve their software. See: https://youtu.be/P3Ljp0MCSzQ?si=umqK9CGv86wiwzKS&t=992