some form of bias is inescapable. ideally i think we would train models on an equal amount of Western/non-Western, etc. texts to get an equal mix of all biases.

Bias is a reflection of real world values. The problem is not with the AI model but with the world we created. Fix the world, ‘fix’ the model.