I was hoping this image might include an adaptive controller, but it’s pretty clear that unless I explicitly specify details like that, the output defaults to “random, normative guy who just happens to be in a wheelchair.” I’ve been intentionally keeping the prompts simple to test what assumptions the model brings to the table—what it imagines when I don’t spell everything out. That simplicity has been a way to surface the underlying biases. Moving forward, I’ll try feeding in more detail to see how close I can get to something I consider more realistic. Of course, that’ll reflect my own biases too—but that’s part of the point.