LLMs are inherently backward looking; their training corpus consists of existing works and their ability to extrapolate is ultimately constrained by that corpus. However powerful their proponents claim them to be, they aren't powerful enough to predict the evolution of human society and culture as a whole.

Changes to that corpus that reflect the real world are going to come from incorporating future works created by humans. (LLM generated training data can reinforce fidelity to the current corpus but don't alter it.) The future still belongs to human creativity.