On: Optimizing LLMs from a Dataset Perspective

[theptip]

Sounds vaguely like the paper “textbooks are all you need”? Though they are not explicitly trying to remove the copyright taint.

https://arxiv.org/abs/2306.11644