フリー問題

問題文

How do you make visual information inside documents reachable by a retrieval-based assistant?

選択肢

1. Store the images in blob storage and reference them by URL only, so the assistant can hand the link to the user and let them look at the figure themselves
2. Generate descriptions of the figures during ingestion and store them as searchable fields in the index alongside the body text of every one of the documents
3. Rely on the model to infer what the figures show from the surrounding text, since a figure in a well-written document is normally explained in the paragraph next to it
4. Send the whole document to a multimodal model on every question so the figures are always read from the original file and nothing depends on what was captured at ingestion