Camera ready complete & available!
β Emily M. Bender (@emilymbender) January 23, 2021
On the Dangers of Stochastic Parrots: Can Language Models Be Too Big? π¦https://t.co/CSMZPlJd8h
To appear in #FAccT2021 w/@timnitGebru @mcmillan_majora and Shmargaret Shmitchell
cc: @mmitchell_ai
Sad to see that only 37% of the French population sees AI and automation as being "good for society."
β Yann LeCun (@ylecun) January 22, 2021
It is "very French" to see the worst in everything, but one can't miss out on the next industrial revolution.
Thankfully, young people are much more positive than their elders. https://t.co/81coVJoCPg
Introducing an improved system for automated transliteration of place names in Google Maps that yields high accuracy and wide coverage for 10 different Indian languages, greatly improving search and discovery for Indian language users. Learn more below: https://t.co/mTGbYmxTIq
β Google AI (@GoogleAI) January 22, 2021
Situation: The AUC is 0.65.
β Karandeep Singh (@kdpsinghlab) January 22, 2021
Clinician: But you left out this key variable! In our analyses, it has a p-value of 0.001 with an odds ratio of 1.2, a pretty important predictor!! No wonder the AUC isn't higher.
*Variable added*
Situation: The AUC is 0.65.
Extremely encouraging to see causal graphs trickling into NLP, enforcing my speculation that " learning from the way causal reasoning was domesticated, would benefit researchers in other area of AI, including vision and NLP."
β Judea Pearl (@yudapearl) January 22, 2021
(in https://t.co/GCeYu5nEol). https://t.co/GP9xHQGhGd
Normalizer-Free ResNets: Our ICLR2021 paper w/ @sohamde_& @SamuelMLSmith
β Andy Brock (@ajmooch) January 22, 2021
We show how to train deep ResNets w/o *any* normalization to ImageNet test accuracies competitive with ResNets, and EfficientNets at a range of FLOP budgets, while training faster.https://t.co/2WMhCkaxJh pic.twitter.com/nwF7lT25BK
π₯ Coming soon to @spacy_io v3 and https://t.co/j3J6mQr8AN π₯
β Ines Montani γ°οΈ (@_inesmontani) January 22, 2021
βΎοΈ span categorizer component to predict labels for overlapping spans & phrases
βΎοΈ annotation mode for overlapping spans
βΎοΈ mix with classic NER or custom suggestion logic
βΎοΈ efficient Doc.spans data structure pic.twitter.com/ZRTCRua9fJ
Over 700,000 paintings from the Rijksmuseum are being released under Creative Commons 1.0 Universal (CC0 1.0) Public Domain Dedication https://t.co/4DiGP6Hu0j pic.twitter.com/QPBg8aYyN9
β Amber Case (@caseorganic) January 22, 2021
Introducing Room-Across-Room (RxR) β the first multilingual dataset for vision-and-language navigation β and the associated RxR challenge, which encourages the ML community to train and evaluate their own instruction following agents. Learn more at https://t.co/hHGiPtmdSl pic.twitter.com/ex3LkxCuX2
β Google AI (@GoogleAI) January 21, 2021
Getting a job is an arms race of algorithms, and not the only one: think about the essay grading algorithms and the essay producing ones! https://t.co/cjz9B1tqwR
β Cathy O'Neil (@mathbabedotorg) January 21, 2021
Me at work: "Remember to use just enough complexity to get it done."
β Chris Albon (@chrisalbon) January 21, 2021
My side projects: pic.twitter.com/LXIdCMGoUu
dplyr's killer app is that you can write code that generates SQL, so you don't have to switch your approach halfway through an analysis
β David Robinson (@drob) January 21, 2021
The siuba package brings this to Python! https://t.co/3egEb1b6QY@chowthedog #rstudioglobal pic.twitter.com/BYis4dkIiS
