We're likely seeing a similar "illusion of understanding" with AI's latest "reasoning" models, and seeing how that illusion can break when the model runs in to unexpected situations.