Amplifying time-awareness features in LLMs via sparse autoencoders reduces look-ahead bias in forecasting while preserving general performance.
Caution ahead: Numerical reasoning and look-ahead bias in ai models
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Forecasting With LLMs: Improved Generalization Through Feature Steering
Amplifying time-awareness features in LLMs via sparse autoencoders reduces look-ahead bias in forecasting while preserving general performance.