> Trying to go straight for the goal is risky.

That's the point of it. You need to take more risk for different approach. Same as what OpenAI did initially.