AI-powered
podcast player
Listen to all your favourite podcasts with AI-powered features
Are We Sitting Too Low When We Think About Our Human Ability as the Baseline?
When you think about machine learning, AI, especially in terms of computer vision, are we being unambitious when we set the baseline at the human's ability? I think we're sitting too low if we look at single image. It's very theoretical to worry about accuracy against what can be seen in an image. If you have someone using your data, they don't give you any prizes for saying, oh it's hard to spot in that particular image. They don't care.