Table 4 Comparison of version 3.5 and 4 of ChatGPT language models with varying runs and threshold settings, illustrating the impact on low and high accuracy metrics
From: Explanatory argumentation in natural language for correct and incorrect medical diagnoses
LM | Run | Threshold | Low acc. | High acc. |
|---|---|---|---|---|
v-3.5 | 3 | 20% | 0.64 | 0.73 |
3 | 50% | 0.79 | 0.82 | |
5 | 20% | 0.64 | 0.79 | |
5 | 50% | 0.78 | 0.84 | |
v-4 | 3 | 20% | 0.63 | 0.72 |
3 | 50% | 0.8 | 0.82 | |
5 | 20% | 0.63 | 0.73 | |
5 | 50% | 0.8 | 0.81 |