From c18f5f0d7674371edfac2b6f5815d72e2608b764 Mon Sep 17 00:00:00 2001 From: Michael Wornow Date: Tue, 18 Feb 2025 21:59:25 -0800 Subject: [PATCH] Update index.html --- index.html | 7 ++++--- 1 file changed, 4 insertions(+), 3 deletions(-) diff --git a/index.html b/index.html index 3de50237..8a401ae2 100644 --- a/index.html +++ b/index.html @@ -77,7 +77,7 @@

Zero-Shot Clinical Trial Patient Matchi
- Stanford University
Under Review + Stanford University
NEJM AI 2024
*Indicates Equal Contribution
@@ -262,12 +262,13 @@


Interpretability

The results show that GPT-4 is able to provide legitimate rationales for most its decisions. When GPT-4 makes a correct eligibility decision (Figure 4), 89% of its rationales were judged as fully correct, 8% as partially correct, and 3% as incorrect. When GPT-4 made an incorrect eligibility decision (Figure 5), its rationales were split 67/8/25%.

- Figure 4: Clinician assessment of whether the rationales generated by GPT-4 given eligibility decision is evaluated as correct. + Figure 4a (top): Clinician assessment of the rationales generated by GPT-4 for its correct eligibility decisions.
+ Figure 4b (bottom): Clinician assessment of the rationales generated by GPT-4 for its incorrect eligibility decisions.

- MY ALT TEXT + clinician_rationale.png