To what extent should we trust AI models when they extrapolate?

Open in new window