Reasoning before code generation lowers the average bias rate in LLM-generated code from 0.64 to 0.40, and a new probe-based method (ProbeDebias) cuts bias by about 84% while keeping quality nearly unchanged.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.SE 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
How Reasoning Shapes Social Bias in LLM-Generated Code?
Reasoning before code generation lowers the average bias rate in LLM-generated code from 0.64 to 0.40, and a new probe-based method (ProbeDebias) cuts bias by about 84% while keeping quality nearly unchanged.