Fine-tuning on data aligned with an LLM's prior knowledge induces overconfidence, and CogCalib mitigates this by gating a calibration loss to known data.
Title resolution pending
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
-
Towards Objective Fine-tuning: How LLMs' Prior Knowledge Causes Potential Poor Calibration?
Fine-tuning on data aligned with an LLM's prior knowledge induces overconfidence, and CogCalib mitigates this by gating a calibration loss to known data.