To clarify, there is not a precomputed DB of an enormous number of questions although this would be possible to derive and does occur on a lower level for caching performance purposes. The total count comes from permutation/probability maths based on the question construction algorithm -- when you request a question it generates one which means I can extend the pool quite easily without re-generating a monster cache table.
It is impressive how good Wolfram is at decoding logic, I'll have to have to think about the question construction but I can't make the question too confusing for a real person to solve. As someone mentioned, maybe more abstract questions would be stronger but the difficultly of course lies in generating them. I certainly think logic questions are weaker than a decently obscured/randomised image captcha, but they come with other advantages and work in text-only contexts (e.g. IM-type challenges).
To clarify, there is not a precomputed DB of an enormous number of questions although this would be possible to derive and does occur on a lower level for caching performance purposes. The total count comes from permutation/probability maths based on the question construction algorithm -- when you request a question it generates one which means I can extend the pool quite easily without re-generating a monster cache table.
It is impressive how good Wolfram is at decoding logic, I'll have to have to think about the question construction but I can't make the question too confusing for a real person to solve. As someone mentioned, maybe more abstract questions would be stronger but the difficultly of course lies in generating them. I certainly think logic questions are weaker than a decently obscured/randomised image captcha, but they come with other advantages and work in text-only contexts (e.g. IM-type challenges).