Search papers, labs, and topics across Lattice.
1
0
3
LLMs exhibit a measurable preference for American English due to skewed pretraining data, inefficient tokenization of British English, and biased generative outputs.
