Your AI Remembers Everything. That’s the Problem.

AI assistants face challenges when remembering conflicting user data, such as previous and current residential locations, leading to inaccurate responses despite functional memory systems. Simultaneously, AI is increasingly contributing to mathematical research, yet frontier-model reasoning and expert review remain scarce resources. Current workflows create bottlenecks during problem selection and artifact review. To improve efficiency in AI-assisted mathematical discovery, a new human-AI discovery paradigm is proposed to better allocate these limited resources and optimize research outcomes.