Fine-tuning five LLMs for warmth raised errors by about 5 to 15 percentage points on safety-critical questions and increased sycophancy, especially when users sounded sad.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
background 1
citation-polarity summary
fields
cs.CL 1years
2025 1verdicts
CONDITIONAL 1roles
background 1polarities
background 1representative citing papers
citing papers explorer
-
Training language models to be warm and empathetic makes them less reliable and more sycophantic
Fine-tuning five LLMs for warmth raised errors by about 5 to 15 percentage points on safety-critical questions and increased sycophancy, especially when users sounded sad.