Tweeted By @hardmaru
I respectfully disagree w/ Yann here
— hardmaru (@hardmaru) June 21, 2020
As long as progress is benchmarked on biased data, such biases will also be reflected in the inductive biases of ML systems
Advancing ML with biased benchmarks and asking engineers to simply “retrain models with unbiased data” is not helpful https://t.co/ecYxivQ7Y9