to 
No, I will not make any such change, for two reasons:https://twitter.com/yoavgo/status/1353413967889244161 …
Pokaż ten wątek
-
As for the claim that our paper is one-sided, this is exhausting. All of ML gets to write papers that talk up the benefits of the tech without mentioning any risks (at least until 2020 w/broader impact statements), but when a paper focuses on the risks, it's "one-sided"? >>
Pokaż ten wątek
![Screen cap of part of sec 4.4 of the paper, reading: "When we rely on ever larger datasets we risk incurring documentation debt, i.e. putting ourselves in a situation where the datasets are both undocumented and too large to document post hoc. While documentation allows for potential accountability [13, 52, 86], undocumented training data perpetuates harm without recourse. Without documentation, one cannot try to understand training data characteristics in order to mitigate some of these attested issues or even unknown ones. The solution, we propose, is to budget for"](https://pbs.twimg.com/media/EshpGqXVcAAktqL.png)
It was really refreshing to read a paper about the risks of large scale NLP models. We always hear about the dangers and bias with facial recognition and image processing, but there isn’t enough discussion about NLP. It’s not something I had properly consider before.
