Is an MCQ really that useful of an interface? The heavy lifting is the computation done around it.
I think just MCQs are not the full shape yet, there is probably something else and something better, maybe different things for different shapes. E.g. a slider, a toggle, etc etc. These are effectively logical rules, there is so much overlap here to the interfaces demo (how do we visually distinguish different parts of an idea).
So don't get too caught up in the exact mechanism of a MCQ, it could have the benefit of just reduced noise to the user rather than the means itself being the most useful. E.g. it could be inheriting the benefit of a more robust system with better memory etc.
Why is a list of options the most ideal?
- Limits decision surface area (well the model is already doing a lot of the deciding, that is an assumption and could also be true for other interfaces).
I think I need to be careful here not to conflate a superior predictive system, one that puts immense effort into modeling the user's wants, and the ideal interface.
See Induced fit.
Smarter models
I do wonder how much the ideal shape is just a bunch of Claude skills and checks, trying to build the UI around it feels a bit premature.
See Grown, not refined.