Follow @avatar42 on Micro.blog.
Great talk on using system prompts to improve LLM responses and reduce token usage.