Language models fail at balanced parentheses because unreliable internal components that promote wrong tokens can outvote reliable ones, and amplifying reliable components fixes the errors.
Mechanistic interpretability for AI safety - a re- view.Transactions on Machine Learning Research, 2024
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Failure by Interference: Language Models Make Balanced Parentheses Errors When Faulty Mechanisms Overshadow Sound Ones
Language models fail at balanced parentheses because unreliable internal components that promote wrong tokens can outvote reliable ones, and amplifying reliable components fixes the errors.