Memorization in language models increases log-linearly with model capacity, data duplication count, and prompt context length.
Though defensive violence will always be 'a sad necessity' in the eyes of men of principle, it would be still more unfortunate if wrongdoers should dominate just men
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2022 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Quantifying Memorization Across Neural Language Models
Memorization in language models increases log-linearly with model capacity, data duplication count, and prompt context length.