Are You A "Copy-Paster" Or A "Summarizer"?
Or something else entirely?
This is a brief report combining a qualitative assessment of LLM style with some objective data from two RCTs of LLM use – using these two sources to create a typology of clinicians in their interactions with LLMs.
The structure of the study is fairly straightforward:
And the results are obviously limited by the cohort and trials involved, but have gross face validity. They describe four styles:
“Copy-paster”: simply takes the unedited raw clinical data or vignette and pastes it into an LLM.
“Selective copy-paster”: curates the clinical content or vignette with the goal of feeding the LLM the most salient sections
“Summarizer”: creates their own summary of the clinical data or vignette in the LLM
“Searcher”: uses the LLM like a search engine, with queries based off the clinical data or vignette
Sounds fairly reasonable – but what would be further interesting to see from this initial work would be the impact of LLM style on results. It is not yet well-described whether a certain style is more likely to generate a more accurate or usable output, but this typology might provide a framework to explore this question.