I am inclined to agree, i believe that once society is able to fill everyone's needs and everyone can summon any ai vr experience they want crime will stop to exist, there would be nothing to gain from committing harm. But i fear the simulated role-play in the context of psychological torture, csam could lead to making dangerous people more confident before we get to that post-scarcity. Maybe you say chatgpt inst realistic enough for it now, but i will be soon.

training an LLM entirely by yourself with self curated text is beyond what is feasible, most ai researched today dont even know whats in all of the data they use. Its more then you can look at even with an extended lifetime and at best you can fine-tune a standard base model.