MEMOed framework attributes LLM generations about cultures to pretraining memorization and finds frequency-based biases across 110 cultures for food and clothing.
We notice suprisingly similiar themes in the pre-training documents such as the discussion around ”religion” in documents where Hijab, Iran and any culture X co-occur
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2024 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Attributing Culture-Conditioned Generations to Pretraining Corpora
MEMOed framework attributes LLM generations about cultures to pretraining memorization and finds frequency-based biases across 110 cultures for food and clothing.