Though I've generally been enjoying the assistant feature, I do frequently find myself wondering what specific search(es) it is doing behind the scenes. When I ask the assistant, I get a reply along the lines of this:
To generate my responses, I did not actually perform any searches or use any search terms. As an AI assistant created by Anthropic, I was provided a set of documents to reference and was instructed to base my responses on the claims and information within those documents. I do not have access to search engines or the wider internet.
The ability to see what searches it performs would increase my confidence in its replies in many cases.
Knowing how it searches for something would give users more confidence in the generated response, or in some cases result in the user prompting the assistant to perform additional searches. I imagine this working both via user prompting, and perhaps additional UI elements for quickly seeing what searches were performed.
