How do I use web content to inform my results?

You can incorporate external sources of information to provide specific context and thereby improve the accuracy and relevance of your model’s results. This is called RAG or “Retrieval-Augmented Generation”, a process to optimize the output of models by using provided data to supplement and enhance their pre-trained knowledge.

You can use information available over the web to inform results returned from Dartmouth Chat. Subsequent feedback from the model will begin by reviewing the web search results, using the context provided to inform the response.

Designating Particular Sites for Context  

One way to use web content is to designate specific websites to be used for context. To add these websites type a “#” into the prompt before putting in the resource’s URL.

In this case, we’re pointing to a particular YouTube video. Dartmouth Chat accesses the text transcript of the video from YouTube and uses it for context on subsequent responses.

Using General Web Site Results  

You can also use the results of a general web search to provide context for your prompt. Subsequent feedback from the model will begin by reviewing the returned web search results, using the context provided to inform the response. To do this:

• Highlight the Web Search icon. To run the web search, click on the Web Search icon and add your search text - e.g., “What is Dartmouth’s policy re: dropping an academic class?”

• Review the results. The model will perform a web search before attempting to address your prompt. The response will include the references utilized to inform the response.

Didn’t find what you needed? Please reach out to research.computing@dartmouth.edu.