LLMs often compute correct answers to simple deterministic tasks in their internal representations, and light fine-tuning of early layers lets them access that information by suppressing a learned prior over outputs.
Under review
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
-
Identifying and Mitigating the Influence of the Prior Distribution in Large Language Models
LLMs often compute correct answers to simple deterministic tasks in their internal representations, and light fine-tuning of early layers lets them access that information by suppressing a learned prior over outputs.