Prototype of using GPT-3 to answer and refine search results (provided by DuckDuckGo), build using Node-RED. This proof of concept is to show how companies like Microsoft and Alphabet can use a language model to provide users of Bing and Google with answer to their questions based on reliable and traceble information. Users can refine their questions or ask for more details in a context-aware way. All sources used to derive an answer are listed at the bottom of the results page.
- Install or start an instance of Node-RED.
- Import the AI Search flow.
- Open the credentials node and enter your own bearer token (in format
Bearer openaiapikeygoeshere, using your own OpenAI key). - Open the AI Search page (by default; http://localhost:1880/aisearch/).
- Type your search query, wait for the response and ask follow-up questions.
This video shows an example of a search with follow-up questions:
housespeaker.mp4
| Topic | Screenshot |
|---|---|
| "Chinese balloon" |  |
| "Earthquake" |  |
| "Tennis" |  |