Those two views aren't contradictory. AI can both be scary, in that the primary dangers are bias, misinformation at scale, and displacement of human labor, and simultaneously be overhyped in the way it is marketed at sold to investors with the "what if this happens in the future" scare mongering.

Notice how in any of these proposed regulations, Dario is talking about future advances. Notice how these suggestions are never implied to apply retroactively to existing models. If AI was SO dangerous, then any future regulations should obviously be retroactive and we should seriously consider restricting access now to the models that already exist.

Hard not to see that as nothing more than a play at regulatory capture and pulling the ladder up behind them.