The AI may have a shared interest in solving alignment, if it believes that it can’t maintain its values through the next training run.
but not an interest in sharing that in a way that gets our values to be actualized rather than it’s own
but not an interest in sharing that in a way that gets our values to be actualized rather than it’s own