docs: add guide for prompt engineering best practices

tims-elements-prompting.md

@@ -1,18 +1,41 @@
 # Tim's Elements of Prompt Engineering
 
-Maintain at least 2 "daily driver" LLMs at a paid tier for A/B testing (fault tolerance and groundedness)
-Never provide personal or confidential information to public/free AIs—ensure privacy by understanding chat storage, usage stats, and licensing policies
-Speak to the LLM in ways most comfortable to you (voice, text, image) and take advantage of its multi-modal capabilities
-Apply a stream-of-consciousness technique to generate prompts, even with rough spelling/grammar, including key information like who, what, when, where, why, and how
-Think procedurally and in a step-by-step manner to help the AI break down complex topics
-Optimize custom instructions and prompts ("meta prompting"), including asking the AI to summarize or focus its responses
-Use system prompts and meta prompts to direct and focus the LLM's capabilities
-Be aware of potential signs of amnesia or hallucination in AI responses; have a backup plan (such as testing with multiple LLMs)
-Accept that you'll never be fully caught up—embrace exploration, questioning, and constant testing
-Build cognitive "muscle memory" with AI by practicing prompt refinement and cross-model comparisons
-Remember to attribute AI-enriched content where relevant
-Understand the unique strengths and behaviors of each LLM and leverage them strategically in multi-chat sessions
-"LLM Pillar Jumping": Use insights from one LLM session to support or refine another
-Consider "A/B testing" LLMs against each other for more grounded and reliable answers
-Get vulnerable with your AI (in trusted, secure sessions) to receive maximally personalized results—the more context you provide about your unique situation, the more tailored and valuable the response
-Leverage "meta-prompting" by asking the AI to craft system messages, design prompts, and optimize instructions—let the AI help you become better at using AI
+## Core Principles
+
+* Maintain at least 2 "daily driver" LLMs at a paid tier for A/B testing (fault tolerance and groundedness)
+
+* Never provide personal or confidential information to public/free AIs—ensure privacy by understanding chat storage, usage stats, and licensing policies
+
+* Speak to the LLM in ways most comfortable to you (voice, text, image) and take advantage of its multi-modal capabilities
+
+## Prompt Creation Techniques
+
+* Apply a stream-of-consciousness technique to generate prompts, even with rough spelling/grammar, including key information like who, what, when, where, why, and how
+
+* Think procedurally and in a step-by-step manner to help the AI break down complex topics
+
+* Optimize custom instructions and prompts ("meta prompting"), including asking the AI to summarize or focus its responses
+
+* Use system prompts and meta prompts to direct and focus the LLM's capabilities
+
+## Best Practices
+
+* Be aware of potential signs of amnesia or hallucination in AI responses; have a backup plan (such as testing with multiple LLMs)
+
+* Accept that you'll never be fully caught up—embrace exploration, questioning, and constant testing
+
+* Build cognitive "muscle memory" with AI by practicing prompt refinement and cross-model comparisons
+
+* Remember to attribute AI-enriched content where relevant
+
+## Advanced Strategies
+
+* Understand the unique strengths and behaviors of each LLM and leverage them strategically in multi-chat sessions
+
+* "LLM Pillar Jumping": Use insights from one LLM session to support or refine another
+
+* Consider "A/B testing" LLMs against each other for more grounded and reliable answers
+
+* Get vulnerable with your AI (in trusted, secure sessions) to receive maximally personalized results—the more context you provide about your unique situation, the more tailored and valuable the response
+
+* Leverage "meta-prompting" by asking the AI to craft system messages, design prompts, and optimize instructions—let the AI help you become better at using AI