ID_AA_Carmack
@qhzq282068
It would be interesting if LLM training tracked the per-token loss back to the source material — it would be an objective measure of how much each specific book / document contributed to the training. Might say something useful for human learning!
0 reply
0 recast
0 reaction