I think writing scenario-style forecasts like AI 2027 is good, actually.

Daniel Kokotajlo and I talk about how they made AI 2027, forecasting vs. science fiction, answers to common critiques (the illusion of certainty, self-fulfilling prophecies), and more: