Utilizing multiple samples when querying a language model for recommendations provides a broader and more comprehensive understanding of the options. Language models are non-deterministic, and asking broad questions yields slightly different results each time. By asking broader questions and comparing multiple outputs, overlapping sets and descriptions can be identified to determine the most commonly mentioned options. This approach allows for last-mile filtering based on personal preferences, providing a more tailored and reliable set of recommendations.
TL;DR: Today we’re releasing a new episode of our podcast How Do You Use ChatGPT? I go in-depth with Notion research engineer Linus Lee on how he uses ChatGPT and Notion AI to maximize creative control. Watch on X, YouTube, or Spotify.
You might think that being an AI researcher would mostly involve solving complicated programming problems and thinking through mathematical equations. Instead, a big part of the job is rewriting parts of your prompts in ALL CAPS in order to make sure the AI model you’re working with follows your directions. “All caps works!” Linus Lee told me in this interview. “If you look at OpenAI's system prompts for a lot of their tools, all caps works.”
Linus is a research engineer at Notion who works on its AI team, prototyping new experiences, like a Q&A chatbot. He is a deep thinker who is obsessed with building AI that enables human creativity and agency. He came on the show to talk about how AI might augment our thinking, how he thinks about prompting to get the best results, and how he uses ChatGPT and Notion AI in his work and life.
I first interviewed him a year ago, when he showed off dozens of AI prototypes he’d been building to try to understand the future of this technology. Our latest interview is a mixture of theory and practice. Linus talks about how the tools we use shape the work we can create and what the future of AI-driven interfaces might be. We watch him demo personal tools he’s built, like an AI chatbot that he communicates with over iMessage. And we peek over his shoulder to see his chats with ChatGPT to understand how he talks to it to get the best results.
Here’s a taste of what we talk about. Read on for more analysis from me at the bottom.
- Using AI to maximize agency. Linus talks a lot about the ways our tools shape our agency as thinkers and creatives—and how AI might be used to enhance rather than reduce our agency.
- AI as a “thought calculator.” Linus borrows a phrase from the popular tech blogger Simon Willison to illustrate dueling points of view on the ultimate goal of AI: is it meant to be a simulacrum of humans or a “thought calculator,” a way to enhance human imagination and creativity?
- Personal prototypes he’s built. Linus regularly experiments with AI on the weekend. He shows us a chatbot he built that works over iMessage, and a new interface for image generators that gives him much better control over their output.
- Better prompting. We go over simple yet powerful techniques for getting the best answer out of AI models—like starting with general queries first, and repeatedly asking the model to answer the same question.
- Using AI for vibe checks. AI is great for reflecting the vibes of books, people, places—and even files on your computer. Linus talks about how he uses ChatGPT to get quick vibe checks that allow him to make decisions.
- Book recommendations. We pit ChatGPT head-to-head against Notion AI to see which can best capture our reading taste. And just when ChatGPT seems like it’s coming out on top, Linus makes a convincing case for Notion AI’s special skill set as an organizational tool that already knows how its users work.
What do you use ChatGPT for? Have you found any interesting or surprising use cases? We want to hear from you—and we might even interview you. Reply here to talk to me!
Miss an episode? Catch up on my recent conversations with writer Nat Eliason and Gumroad CEO Sahil Lavingia and learn how they use ChatGPT.
