Going faster at the expense of increasing tail risk in the future isn't a gain in efficiency. It might even be a net negative considering how bad the risky outcomes get.

To gain efficiency, you have to find a way to go faster while also keeping your risks the same or less than before. Otherwise you're just building increasingly fragile things at a faster and faster rate.

Any conversation about the efficiency gains AI brings has to take this into account or the whole enterprise is dishonest.