Heatmap indicates balanced accuracy score for each NGA (bottom label) and UN geographic region (top label) over time (left axis) for the GPT-4 Turbo model (IMAGE)
Caption
The figure shows the balanced accuracy distribution across space and time for GPT-4 Turbo, the model with the best overall performance. Darker colors indicate greater balanced accuracy, while completely white areas signify the absence of data points. More recent periods are generally colored lighter, indicating lower accuracy of the model. Although one might assume that lower accuracy in more recent periods is due to more data being available, this is not necessarily true. As an example, the model's accuracy is higher for the earlier years of the NGA Basin of Mexico, where there are roughly the same number of data points between 5000 BCE and 1000 BCE.
Credit
Complexity Science Hub
Usage Restrictions
Please mention with credit
License
Original content