Untitled

I'm still not sure how we could claim alignment if we (a) don't know what happens during inference (2) can't define intelligence and (d) are limited to making polite requests via markdown.