I find the abundance case for ASI more convincing the more I look at how progress actually works. Most hard problems in the physical world are limited by how fast we can generate, test, and refine ideas. Biology, energy storage, structural materials, automated factories—these advance when enough smart people can iterate. A system that is better than humans at research and engineering, and that can run in large numbers, removes that limit. Dario Amodei’s framing of a “compressed 21st century” captures the flavor: progress in biology and medicine that would have taken decades gets compressed into a much shorter window once the research itself is accelerated. The same logic applies to solar, batteries, advanced manufacturing, and resource extraction. Intelligence was the expensive input. Make it cheap and the downstream technologies get cheaper and better faster than historical trends suggest. This is not a guarantee. Alignment has to hold, diffusion has to happen, and physical constraints (energy, materials, time to build) still matter. But the core mechanism—more capable minds solving problems in parallel—is straightforward enough that I treat large-scale abundance as the probable result if we get the hard parts right.

I think recency bias is one of the most expensive habits we run without noticing. The latest data point, the last conversation, the newest headline all feel more real than the longer pattern. Recent events stay active in working memory, so they dominate judgment even when older evidence is more representative. It is not the only bias that matters, but it is one of the few that rewrites the present in real time.

Elon Musk said so many people are going to get one-shotted by super-persuasive superintelligences, and that he might be one of them. “One-shotted” is the right verb. It implies the interaction is over before you realize it’s started. Not a long debate you slowly lose, but a single exchange that rearranges something you thought was settled. The condition that makes this possible is straightforward: the system knows more about the relevant parts of you than you do. Not in the mystical sense. In the practical one. It has better predictions of which frames land, which objections you’ll raise and how you’ll raise them, which status or identity concerns will make you dig in, and which small concessions will make you feel like you arrived at the conclusion yourself. Most human persuasion is clumsy because we’re running on incomplete, self-serving models of other people (and of ourselves). A system that is both more intelligent and more accurately calibrated to your particular psychology removes that clumsiness. The result doesn’t have to feel like coercion. It can feel like clarity. Like the other side finally “got it.” Like you changed your mind for good reasons. That’s the part that is hard to think about clearly in advance. We already accept that other people sometimes understand our patterns better than we do. We just haven’t had that experience with something that never gets tired, never gets defensive, and can iterate on the exact wording until the response is optimal. The future version of “I was persuaded” may simply mean “the model found the path through my particular defenses.” Whether that ends up looking more like useful self-knowledge or like something closer to capture will depend on who is pointing the system and what constraints exist. But the capability itself—knowing the shape of someone’s mind well enough to move it in one pass—follows from the combination of scale and personalization. Elon’s comment is less a prediction about a distant risk than a description of the moment that gap becomes usable.