Titans 在推理时学习：如果它学错了怎么办？

Titans gives AI a separate neural memory that updates its own weights as information arrives.

I wrote a short visual explainer on how it works. In my small synthetic experiment, correcting false inputs improved recall—but didn’t fully restore it.

What should an AI be allowed to remember automatically—and how should we undo a mistake?