LLMs show up to 60.58% social bias in generated code; a new Fairness Monitor Agent cuts bias by 65.1% and raises functional correctness from 75.80% to 83.97%.
arXiv preprint arXiv:2005.00268 , year=
2 Pith papers cite this work. Polarity classification is still indexing.
2
Pith papers citing it
verdicts
UNVERDICTED 2representative citing papers
Deduplicating training datasets reduces language model verbatim memorization by 10x, improves training efficiency, and enables more accurate evaluation by cutting train-test overlap.
citing papers explorer
-
Social Bias in LLM-Generated Code: Benchmark and Mitigation
LLMs show up to 60.58% social bias in generated code; a new Fairness Monitor Agent cuts bias by 65.1% and raises functional correctness from 75.80% to 83.97%.
-
Deduplicating Training Data Makes Language Models Better
Deduplicating training datasets reduces language model verbatim memorization by 10x, improves training efficiency, and enables more accurate evaluation by cutting train-test overlap.