By re-sampling alternate tokens at each position of generated text, the authors find many 'forking tokens' where a single token choice flips the model's final answer distribution, including chain-of-thought steps and punctuation-like tokens.
Answer: Terry loves to swim but he never learned how to dive.He would watch his friends gracefully dive into the pool while he awkwardly jumped in
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
dataset 1
citation-polarity summary
fields
cs.CL 1years
2024 1verdicts
CONDITIONAL 1roles
dataset 1polarities
use dataset 1representative citing papers
citing papers explorer
-
Forking Paths in Neural Text Generation
By re-sampling alternate tokens at each position of generated text, the authors find many 'forking tokens' where a single token choice flips the model's final answer distribution, including chain-of-thought steps and punctuation-like tokens.