Almanac
paper

Forecasting With LLMs: Improved Generalization Through Feature Steering

paperactiveprovisionalforecasting-with-llms-improved-generalization-through-feature-steering-8d6e4f01·1 events·first seen 2d ago

Aliases: Forecasting With LLMs: Improved Generalization Through Feature Steering

Co-occurring entities

More like this (12)

Recent events (1)

5arXiv · cs.CL·2d ago·source ↗

Feature steering via sparse autoencoders reduces look-ahead bias in LLM forecasting

Researchers apply sparse autoencoders to inspect LLM internal states during forecasting tasks, identifying features associated with time-aware versus look-ahead-biased reasoning. Amplifying time-awareness features causally reduces look-ahead bias while preserving general reasoning performance, whereas directly steering look-ahead-bias features has no effect. The work demonstrates that interpretable temporal features can shift LLMs toward more historically grounded forecasting. This is a mechanistic interpretability result with practical implications for LLM-based prediction systems.