Seonglae ChoAI Bias Benchmarks
AI Bias Notion
Clearly defining "what constitutes bias" across 5 dimensions, enabling control in both training and monitoring. During GPT-5 training, a self-evaluation model (LLM grader) scores responses → reflected as a bias loss term. This is not simply "data cleansing" but a structural improvement that incorporates bias itself as a learning objective
Defining and evaluating political bias in LLMs
Learn how OpenAI evaluates political bias in ChatGPT through new real-world testing methods that improve objectivity and reduce bias.
https://openai.com/index/defining-and-evaluating-political-bias-in-llms/
Equality of Outcome & Equality of Opportunity
Binary Classification — holisticai documentation
The idea is that an unbiased system would present roughly similar success rates across groups.
https://holisticai.readthedocs.io/en/latest/getting_started/bias/metrics/binary_classification.html