This replay of our May 2025 members’ meeting features AI expert Curtis Michelson of Jedi on the Fly. Michelson guided participants through advanced strategies for making generative AI more useful, creative, and impactful in innovation work. Slides and polls from the session are available here. Video replay is below.
The session focused on how to enhance prompt effectiveness, analyze structured and unstructured data using LLMs, experiment with a technique called “vibe coding” for prototyping, and apply “metaprompting” to generate better AI inputs. Designed for innovators and strategists, the session offered live demonstrations, practical frameworks, and tools to help users deepen their AI intuition and become more intentional partners with generative technologies.
Key Themes and Tips
- Generative AI is a Collaborative Partner, Not a Replacement
Michelson emphasized that LLMs (like ChatGPT and Claude) should be seen as creative collaborators, not tools for outsourcing human insight. His approach centers on co-creating with AI while maintaining authorship and intellectual control. - Prompt Enhancements: Use “Twists” to Supercharge Output
He introduced practical prompt enhancers (“twists”) such as:- ELI10 – “Explain like I’m 10” to simplify technical content.
- MECE – Ask for responses that are “Mutually Exclusive, Collectively Exhaustive” to improve structure and clarity.
- Structured + Unstructured Data as LLM Fuel
Michelson demonstrated how LLMs can ingest and analyze both structured (e.g., spreadsheets) and unstructured (e.g., email exports) data. He showed how to use these models to extract insights and generate visualizations from datasets locked inside inboxes or newsletters. - Vibe Coding: Building Prototypes On-the-Fly
“Vibe coding,” a term coined by Andrej Karpathy, refers to using LLMs to iteratively build prototypes and dashboards in real time. Michelson showed how he and collaborator Hamid created a fully functioning dashboard from GenAI use case data with minimal code, using Claude for its strength in structure and presentation. - Metaprompting: Designing Better Prompts with Prompts
He introduced a tool called Jedi on the Fly, designed to help users create better prompts via “metaprompting.” This includes:- Clarifier – Refines and improves vague ideas.
- Instructor – Converts ideas into detailed, structured prompts.
- Colliders – Applies thinking models (e.g., Six Thinking Hats, TRIZ) to stimulate more innovative perspectives.
Michelson noted this method leads to better research prompts, ideation sessions, and writing scaffolds.
- Customization Through Profile Settings
Michelson shared how modifying ChatGPT’s custom instructions (e.g., interests, tone, and response style) significantly improves relevance and output quality. For example, adding preferences like “terse,” “factual,” or “innovation-focused” helps steer the AI. - Encourages AI Play and Experimentation
He encouraged time for “creative exploration,” such as generating surrealist poetry or unconventional outputs, as a way to spark insights and develop stronger AI intuition. “Weird angles” can lead to useful breakthroughs in innovation settings.
