Comment by Atotalnoob

Safety techniques require you to understand your product and have deep observability.

This and safety techniques themselves can improve the performance of the hypothetical AGI.

RLHF was originally an alignment tool, but it improves llms significantly