The Benchmarking of Machine Learning

i always like to do things that are maybe a bit different tan what people around mere are doing, cause i think i learn more from that. I often feel like benchmarks these days are a lagging indicator. New capabilities happen, and then only after that can you build a bench mark around it. Un like solving new problems, cause my experience has been that it's kind of hit or amiss. Sometimes you get through reviewers who don't understand their problemyuare trying o s at. But it's very high. Verianc ifernce has been like in the last couple of conferences, if yeu have gone lucky.