Community questions the mainstream AI's ideological bias, sparking a discussion on "training bias"

BlockBeats news, May 4: AI community user “X Freeze” posted that mainstream AI models, including ChatGPT, Claude, and Gemini, “less often endorse conservative positions” on topics such as gender, immigration, and crime, and questioned whether their value orientations may have systematic bias.

The view holds that as AI capabilities rapidly improve, its “value alignment” (alignment) process may be affected by training data and design mechanisms, leading to a consistent tendency in certain public issues. The related remarks sparked discussions in the AI community about “training data bias” and “model design orientation.”

At present, major AI development organizations generally say that their model training objectives are to improve information accuracy and safety, and to reduce bias through diverse data and evaluation mechanisms, but controversy over AI value neutrality continues.