Summary of "스탠포드가 가르치는 AI시대 창의력 훈련법 | 스탠포드 교수 제레미 어틀리"
Overview
Jeremy Utley (Stanford adjunct professor of creativity and AI) explains how to get real creative and productivity gains from generative AI by changing how you treat and use it. Key themes:
- Treat AI as a teammate, not just a tool.
- Use voice input and iterative conversation rather than a single typed query.
- Have AI ask questions and roleplay.
- Focus AI on tasks you dread and bring your unique experience/inspiration to get differentiated outputs.
Utley provides practical prompts and examples (including a National Park Service case that scaled huge savings) and cites research showing a large “realization gap”: many people don’t get AI’s potential because they use it the wrong way.
Main ideas and lessons
- Orientation matters: people who treat AI like a teammate (coaching it, iterating, getting it to ask questions) get far better creative and productivity outcomes than those who treat it like a basic tool.
- Voice > typing: speaking to an LLM lets you ramble and offload synthesis to the model, unlocking richer outputs; treating an LLM like a Google search box limits its capability.
- Ask the AI to be an expert and to interview you one question at a time to surface non-obvious ways to apply AI in your work.
- Use AI to automate parts of work you dread. Small, low‑technical experiments can create outsized impact across organizations.
- AI can roleplay (simulate a conversation partner and give feedback) and produce psychological profiles, rehearsals, and critiques — practical “drills” to prepare for real interactions.
- Creativity’s definition doesn’t change with AI: creativity is “doing more than the first thing you think of.” With AI it’s easier to reach “good enough,” so to be exceptional you must pursue volume and variation of prompts/outputs.
- Everyone can be creative with AI; human inputs (experience, perspective, inspiration) produce unique outputs from a shared model.
“Creativity is doing more than the first thing you think of.” — quoted from an unnamed seventh grader in Ohio
Concrete methodology and step-by-step prompts
Diagnose how AI can help you
Prompt template:
You’re an AI expert. I want a consultation to figure out where I can best leverage AI in my life/work. Ask me questions, one at a time, until you have enough context about my workflows, responsibilities, KPIs, and objectives to make three obvious recommendations and two non-obvious recommendations for how I could leverage AI.
- Let the model ask and answer iteratively; provide context as requested.
Capture and synthesize a spoken conversation into working output
- Use voice input to recount a conversation or brainstorm session.
- Ask the model to interview you about the conversation (pulling out angles, facts, decisions).
- Request concrete deliverables, e.g.:
- “Convert this into a memo/outline for an article.”
- “Give three framing options.”
Handle mediocre AI output
- Treat the output as a draft from a teammate.
- Provide feedback, coach the model, request iterations, and ask what additional information it needs.
- Ask the model: “What questions do you need from me to improve this?” and answer them.
Prepare for difficult conversations (roleplay drill)
- Tell the model about the conversation partner and context.
- Ask the model to create a psychological profile of the partner.
- Request roleplay of the partner and then ask for feedback from the partner’s perspective on your approach.
Discover high-impact applications in an organization
- Train non-technical staff on basic AI collaboration principles.
- Encourage people to apply AI to repetitive, paperwork-heavy, or dreaded tasks and to share successful tools across teams.
Maximize creative output with AI
- Prompt for volume and variation (many outputs, many framings).
- Iterate and curate — read, sort, and synthesize the variations to surface exceptional ideas.
- Always add your unique inspiration/experience as input to differentiate outputs.
Examples and evidence
- National Park Service case: A ranger (Adam Rymer, Glen Canyon) used a 45‑minute session to build a natural‑language tool that automated paperwork for replacing carpet tiles. That tool was shared and is estimated to save ~7,000 days of human labor across ~430 parks in a year.
- Research findings cited: AI can make people roughly 25% faster, produce about 12% more work, and increase quality ~40%, yet fewer than 10% of professionals are driving meaningful productivity gains — indicating a “realization gap.”
- Observational finding: In studies, some users became less creative when using AI because they treated it incorrectly; outperformers treated AI as a collaborator and coached it.
Practical tips (compact)
- Use voice input (speech-to-text with LLM) to capture messy thoughts and let the model do synthesis.
- Stop treating LLMs like a search box — verbal, iterative prompting unlocks more power.
- Start with parts of your work you dread or repeat; small automations can scale.
- Get the AI to ask you clarifying questions before producing outputs.
- Coach and iterate on outputs rather than discarding mediocre results.
- Bring your unique perspective, experience, and inspiration to the prompts.
- Prompt for quantity and variety when aiming for exceptional creativity.
Concepts and jargon
- Realization gap: the mismatch between AI’s potential and actual realized productivity/creativity gains by workers.
- Tool vs. teammate orientation: behavioral framing that impacts whether users iterate and coach AI.
- Functional fixedness / satisficing (Herbert Simon): cognitive biases that lead us to accept early, “good enough” solutions instead of exploring more options.
Speakers and sources featured
- Jeremy Utley — adjunct professor of creativity and AI at Stanford University (primary speaker).
- Perry Claybond — co‑author/partner (mentioned as co‑author of Idea Flow).
- Adam Rymer — backcountry ranger at Glen Canyon National Park (case study).
- National Park Service — organization involved in the training and case study.
- Glen Canyon National Park — workplace of Adam Rymer (case study context).
- ChatGPT / LLMs — generative AI models discussed and used in examples.
- Google (search box) — example of a familiar interface that biases user behavior.
- Herbert Simon — referenced for the concept of “satisficing.”
- Unnamed seventh grader in Ohio — quoted for a simple definition of creativity.
- Unnamed collaborator — referenced in an example where voice + AI produced an article outline.
Category
Educational
Share this summary
Is the summary off?
If you think the summary is inaccurate, you can reprocess it with the latest model.
Preparing reprocess...