Use demonstrations (safe exploration) — supplies the baseline that anchors → Bounded exploration

These two remedies are presented as a pipeline, not independent alternatives. The Use Demonstrations paragraph ends by noting that a baseline policy learned from expert trajectories can itself define the region exploration is bounded around: "even if further learning is necessary, exploration away from the baseline policy can be limited in magnitude" (concrete-problems, §"Use Demonstrations:", p. 15). Bounded exploration needs some prior notion of a safe region or a known-safe policy to check candidate actions against; without one, there is nothing to bound exploration around. Demonstrations supply exactly that starting point cheaply, from data the designer already needed to collect, rather than requiring a hand-specified safe region or a learned dynamics model built for the purpose. The combination lets an agent both start safer, by imitating a baseline, and stay safer, by exploring only in its vicinity.