AIShould the logic for training on human knowledge also apply to AI distillation?
A post challenges people who defend training large language models on all human knowledge—even paywalled material—to apply the same logic to distilling their models.
TLDR
A post asks why, if all human-created or discovered knowledge is fair game for training large language models, the same logic wouldn't extend to distilling those models. It sharpens the question for people who also regard a model's chain-of-thought reasoning as meaningful knowledge.
Should the logic for training on human knowledge also apply to AI distillation?
A post challenges people who defend training large language models on all human knowledge—even paywalled material—to apply the same logic to distilling their models.
TLDR
A post asks why, if all human-created or discovered knowledge is fair game for training large language models, the same logic wouldn't extend to distilling those models. It sharpens the question for people who also regard a model's chain-of-thought reasoning as meaningful knowledge.