“Szymon Sidor was initially skeptical about AI safety research because there were few concrete problems to work on when OpenAI was founded.”