LLM Axis is a customizable word organization graph using GPT4. It's a way to explore the relationships between words and concepts. The graph axes and word list are fully customizable.

Under the hood, LLM Axis is the frontend for a fancy prompt to GPT-4. It's an exploration of what domain specific interfaces for requesting an LLM can look like.


Inspiration

I think about this tweet a lot. I agree that the final form of working with large language models should not be "sending texts." This graph experiment was a way for me to explore some other forms of interacting with large language models.


Examples

Fruits


Music Genres

Note: Who's to say jazz is less sophisticated than classical?


Political Figures

Note: I don't think Biden or Trump are taller than Obama. This could reveal the model's bias of who is a "bigger" political figure.


Prompt Scaffolding

When you click "Reorganize Words," the app will send a prompt to GPT4 to get the positions for each of the words in the word list.

The prompt expects a JSON format response from GPT4 to render the words in HTML.


Tech Stack

Frontend: React, Tailwind
Backend: Next.js
Machine Learning: OpenAI GPT-4
Hosting: Vercel