AI development should be slowed down in specific application areas, not just in core LLM development