Increasing transparency and trust in language models with OLMoTrace (allenai.org)

"... lets you trace the outputs of language models back to their full, multi-trillion-token training data in real time."