One Model to Learn Them All and AI Is/Isn’t Taking Over the World (by Silly Rabbit)

The narrative around AI has split into two incompatible camps: one declaring it's already reshaping the global economy, the other insisting it remains embryonic. Yet both framings coexist in serious technical and business circles without anyone settling the question. What happens when companies and countries are forced to make trillion-dollar bets on a technology they can't actually evaluate?

• The technical capabilities are advancing faster than anyone predicted, but the gap between what AI can do and what people claim it will do keeps widening. Machines can now learn from 300 million labeled images and apply single models across multiple tasks, yet this creates new problems for the people trying to fund and deploy these systems.
• Investment decisions in AI are being made by people with no actual experience building the systems they're investing in. Allocators are being asked to choose between competing machine learning strategies without the engineering background to assess which bets make sense.
• The real bottleneck isn't technical anymore; it's organizational and epistemic. The question has shifted from "can machines do this?" to "who knows which machines should be doing what and how would anyone verify that choice?"
• Countries and companies are racing to fund AI R&D at scale, not because they're certain of the outcome, but because falling behind carries obvious costs. The dynamic mirrors past technology races where being second meant irrelevance, regardless of whether the first-place strategy was actually optimal.
• If a cancer cure is hidden in thousands of unread clinical studies and AI could theoretically find it, we're now asking whether the systems being built today will be the ones that matter for that discovery. The stakes are clarifying, but the path to actually getting there remains opaque.