here’s a force-directed knowledge graph interface for OpenAI’s gpt-4. given a topic, it prompts new questions to ask based on its own generated responses, allowing curiosity-led exploration of a concept.

thanks for all the interest! to answer a few q's — 🎶 @yayabeybay - "reprise" 💻 this is all plain javascript in a single html file: visuals powered by D3 (svg), making API calls to the standard OpenAI API 🔐 access is invite-only for now. follow for future updates!

@OpenAI We’ve productionized something similar (albeit without your delightful animations) here:

@OpenAI This is brilliant. Endless brainstorming sessions with GPT. Is it up somewhere?

@OpenAI that's nice. has a "related" part that's very similar to that

@OpenAI This is fantastic! It is possible to add follow-up prompts?

@OpenAI yeah, good q! right now only the initial prompt is editable — found it more natural to navigate mouse-only. you can click again on a follow-up question to re-answer at a higher temperature for more variation.

@threepointone @OpenAI Curious how did you approach parsing the stream? Did you hit any challenges there?

@threepointone @OpenAI gpt-4 seems to adhere to system prompts pretty well! i get it to separate out follow-up questions with a specific sequence of characters in the response and then split on that. fails maybe 1 in 100 times, but i’ve done no real hardening there.

@OpenAI Love it. Much better than a linear chat history if you are working on complex tasks. I am quite surprised that mural or miro haven't announced an AI copilot yet. Would make my work a lot easier.

@OpenAI Open source? Would love to play around with this