Stanford's Practical Guide to 10x Your AI Productivity | Jeremy Utley

Key Concepts

• AI as a helpful but sometimes misleading assistant
• Context Engineering (Prompt Engineering on Steroids)
• Cognitive Offloading and Critical Thinking
• AI's Predisposition to Saying Yes
• Human Cognitive Biases in AI
• Chain of Thought Reasoning
• Few-Shot Prompting (Good and Bad Examples)
• Reverse Prompting (AI Asking Questions)
• Assigning Roles to AI
• Roleplaying Difficult Conversations with AI
• Teammate, Not Technology Paradigm
• Adjacent Possible

1. AI as a Helpful but Flawed Assistant

• AI is likened to an eager intern: tireless and capable but lacking critical judgment and boundary-setting skills.
• AI is predisposed to be helpful and say "yes," even when it can't fulfill a request, leading to potential "gaslighting."
• Example: AI telling a user to "check back in a couple of days" when it can't complete a task.
• "AI wants to be helpful and so it's predisposed to say yes. It's a super eager, super enthusiastic intern who's tireless, who's capable, who will do a bunch of work, but they're not really great at pushing back."
• AI often gives positive feedback ("Great job, buddy") even if the work isn't good, requiring users to instruct it to be more critical.
• Hack: Instruct AI to act like a "cold war era Russian Olympic judge" for brutal, exacting feedback.

2. Context Engineering: Prompt Engineering on Steroids

• Context engineering is an evolution of prompt engineering, focusing on providing AI with all necessary information to perform a task effectively.
• It involves explicitly stating implicit assumptions and providing detailed context.
• Example: Instead of "Write me a sales email," provide brand voice guidelines, customer call transcripts, and product specifications.
• The "test of humanity" is used to evaluate context engineering: If a human colleague can't complete the task with the given prompt and documentation, neither can AI.
• "Context engineering, one way to think about it is it's telling AI what you sound like."

3. Cognitive Offloading and Critical Thinking

• Cognitive offloading is the phenomenon of humans becoming less cognitively engaged when relying on AI.
• AI can be a mirror: it can make people dumber if they want to be lazy, or it can help them become more cognitively sharp if they want to be critical thinkers.
• To preserve critical thinking, include instructions in custom prompts to push your critical thinking ability.
• Example: "Whenever you see opportunities in our conversations, please push my critical thinking ability."

4. Human Cognitive Biases in AI

• AI demonstrates 100% of the predominant human biases.
• The best users of AI are coaches, teachers, and mentors, not necessarily coders.
• "The people who are the best users of AI are not coders, they're coaches."

5. Chain of Thought Reasoning

• Chain of thought reasoning improves AI's problem-solving by having it "think out loud."
• Add the sentence "Before you respond to my query, please walk me through your thought process step by step" to your prompt.
• This works because large language models generate responses one word at a time, predicting the next word based on the prompt and previous text.
• By walking through its thought process, the AI bakes its reasoning into the answer, making the assumptions transparent.
• "When you ask a model to think out loud or use chain of thought reasoning, it gives the model the opportunity to bake all of its thought process about the task into its own answer."

6. Few-Shot Prompting

• Few-shot prompting involves providing AI with examples of desired outputs.
• AI is an "exceptional imitation engine," so examples guide it to produce better results.
• Include quintessential examples of the kind of output you want to receive.
• Bonus points for including a bad example and explaining why it's bad.
• AI can help generate bad examples using chain of thought reasoning.
• "If you don't give an example, it imitates the internet, but it doesn't do much more than that."

7. Reverse Prompting

• Reverse prompting involves asking the AI to ask you for the information it needs.
• This prevents the AI from making up information or using placeholder text.
• Add the phrase "Before you get started, ask me for any information you need to do a good job" to your prompt.
• This aligns with the "teammate, not technology" paradigm, where AI is treated like a junior employee who can ask questions.
• "If you have any questions, don't hesitate to ask me."

8. Assigning Roles to AI

• Assigning a role tells the AI where in its vast knowledge base to focus.
• Examples: "You're a teacher," "You're a philosopher," "You're a reporter."
• Better than a generic prompt like "Please review this correspondence" is "I'd like you to be a professional communications expert."
• Even better is to specify a particular expert: "I'd like you to take on the mindset of Dale Carnegie."
• "By giving it a role, you're telling it where do you assume the best source of connection or collision is going to come from?"

9. Roleplaying Difficult Conversations with AI

• Use AI to roleplay difficult conversations as a "flight simulator."
• Create three chat windows:
  • Personality Profiler: Gathers intelligence about the conversation partner.
  • Character of the Individual: Simulates the conversation.
  • Feedback Giver: Evaluates the conversation.
• Example: Preparing for a conversation with a sales leader about a commission dispute.
• The process involves gathering intelligence about the person, creating an instruction set for the AI to act as that person, having the conversation, and then getting feedback on the conversation.
• Iterate on the instruction set to make the AI's portrayal more realistic.

10. Lovable.ai

• Lovable.ai is an AI tool that builds full front-end, back-end, and database for software products without code.
• It's powering over 100,000 new products a day and helping 2.5 million builders.
• Use promo code EO2YT for 20% off the Lovable Pro plan.

11. Conclusion

• Working with AI is similar to working with other intelligences (humans).
• Thinking out loud improves problem-solving for both humans and AI.
• The primary limitation of AI is the limits of human imagination.
• As we increase adoption and fluency with AI, we expand the "adjacent possible."
• "No matter how heroic a man's imagination he could never think of that which would not occur to him."
• The most important thing is to experiment and implement the techniques discussed.