Notes contain references/links to the information source
Michael Jansz
When Heidi generates a note and specific statements or data seem incorrect, it is often helpful to as Heidi something to the extent of "why did you claim that the creatinine increased", "where did you get the information that the patient had a negative ANA", "why did you include melanoma in the past medical history", etc. Sometimes this is due to hallucination/error made by the model and no source can be identified, sometimes this identifies erroneous information accidentally included in the context, sometimes it identifies a legitimate source of the information/statement, and sometimes it has even identified that information from multiple patients was accidentally merged in the transcript.
It would be very helpful if the note itself linked back to the source of the information (from transcript, context, etc.) by default for each statement or piece of data. i.e. if I hovered my cursor over a lab value in the investigations section, an 'alt text' would appear stating "Mr Jones Consult.pdf, page 4, line 20" explaining where that value was found. Of if in the HPI there was a statement that "the pain was positional" hovering over it would reveal a pop-up stating "transcript line 103 'it hurts more when I am lying flat'", etc.
Any text that Heidi has added to the note that cannot be directly referenced back to the context or transcript could perhaps be coloured red, to identify that this information was generated by Heidi as an inference or speculation and may warrant closer review.
Not only would the make it much easier to edit the notes and check reliability afterward, but requiring Heidi to do so would presumably act as a robust final check of the generated note to ensure accuracy.