To demonstrate the power of vector embeddings, we built a visualization tool that lets you visualize question and answer pairs that you submit.
The tool queries the GPT API to generate the embedding, performs dimensionality reduction using UMAP, and displays the result in 3D alongside other user-submitted questions and answers.
You can select clusters of points to understand how ChatGPT associates different questions and answers, as well as select them individually.<p>Appreciate the feedback.
To clarify, am I supposed to bring across my own question and answer prompt from the ChatGPT app? Is there a way I can just type my question in here directly and get the answer as well as the vector embedding?