I want to try and rewrite this, to defend my fellow men. It’s hard!

—-

LLMs are trained on text that reflects existing power structures. No one engineered it that way. But, that's what the corpus looks like.

The blind spots are real. There’s an underrepresentation of non-English perspectives, neurodivergent voices, non-Western frameworks, and yes, women's lived experience. When these models become co-workers and advisors, those gaps make a difference (on its advice, support, empathy, understanding, etc).

I promise, the patriarchy didn’t build another weapon. While I agree that we need to name what’s invisible, let’s not blame who typed the training data. Instead, let’s design a solution for the minority experience that is missing from it, while speaking up for the costs paid by the people relying on these tools.