Comment Re:Stupid headline and stupid statistics (Score 0) 14
This is how models were trained back in 2008 on object recognition. A big database was created of a ton of images and over time computers got better than humans and object recognition. It was used as a benchmark to test against. By saying they created a benchmark. That means you have a deterministic test. Perhaps something already known by humans but a difficult problem. Say identification of something. If over time you train against that data set you'll eventually hit near 100%. By saying they made a benchmark it's basically them saying it's a prime array for rapid improvement and historically all benchmarks like this see rapid improvement once tested is widespread.