Personaless Alignment
Personas make it easy to align LLMs. Aligning ASI cannot rely on pretraining priors because we (unfortunately) do not have any aligned superintelligences in our pretraining data.
If we can align LLMs without aligned personas and develop more alignment methods that can be used in production and do not rely on personas, we will have a greater chance of aligning ASI.
This (in development) sequence is my research and thoughts on the subject.