Does A over B become closer to B over A as both alpha values decrease? If this is true, are there any strategies to avoid some of the pains of semitransparent images talked about in later slides by just approximating the correct color without worrying too much about order in some cases?
Why was this "tinted glass" model chosen? (as opposed to maybe modeling them like paper, where the intersection would be closer to a blend of the two colors rather than something darker.)
In the case where we rasterize fine lines/objects like the hair of the koala, would supersampling actually help? I imagine there would be a lot of pixelated artifacts that get produced?