Current implementation of Chain of Thought visualisation is not real CoT

It is just generation of what the model thinks it CoT is, not the real CoT embedded in <think> </think> Im pretty sure