GPT-4 in the web UI defaults to trying to describe the image very tentatively and in general, and it seems intentional: doing that improves performance if you'r

GPT-4 in the web UI defaults to trying to describe the image very tentatively and in general, and it seems intentional: doing that improves performance if you're expecting follow up questions similarly to chain of thought. I expect the API with low temperature will be much more decisive in describing images