Dan Roy    @roydanroy    11/25/2021      

There is increasingly a cost to asking for more experiments. Computation is expensive and environmentally problematic. Demanding exhaustive comparisons disadvantages smaller laboratories with fewer resources. Is it not embarrassing to burn down a small forest unnecessarily?
    3         93



Stella Rose Biderman    @BlancheMinerva    12/4/2021      

This does highlight to me one of the great tragedies of current trends in NLP research... there are sections of this paper I could have hypothetically written, but didn't have the access to the necessary computational resources to carry out the experiments.
    1         4

Brandon Rohrer    @_brohrer_    15 hours      

You all are great. Thank you for turning this thread into a trove of RL resources.
    1         2

Daniel Whitenack    @dwhitena    11/30/2021      

Being able to read unlocks doors to new aspects of life, from health to income. @SILintl helps communities connect to everything they need to read & write. Local leaders are ready & asking for resources. Give now to help these communities: https://t.co/qbJp12Zuv4 #GivingTuesday
    1         1

Dan Roy    @roydanroy    11/25/2021      

I feel that reviewers who suggest more experiments ought to explain what we might learn from... more experiments. What hypothesis are they testing? As far as I can tell, none. We seem to be suffering as a community from not teaching our students how to do empirical work.

Khimya    @khimya    12/6/2021      

#BehindtheScenes 1st submission at AISTATS20; the paper was fresh from the oven & had no experiments. The theory held only for a strict case of formalism. 2nd revision at ICML21 was rejected. Constructive reviews, major rework on the theory, & experiments led to stronger work 7/N