Top Tweets for #ethnlp

almost 3 years ago

@alan_ritter What did surprise me (unless I’m misreading it?) is the high ratio of papers flagged for ethics review: is it really about 1 in 4?? 🤯 #ethnlp #nlproc

0

1

0

146

over 3 years ago

I’m surprised by this #ethnlp tweet (see entire 🧵↕️). White-hat hackers report weaknesses/exploits to the system/model deployer early, giving them adequate time to fix before going public/OS. I don’t see why at least similar measures shouldn’t be taken here

Andy Zou

@andyzou_jiaming

over 3 years ago

So why did we publish it? Despite the risks, we believe it to be proper to disclose in full. The attacks presented here are simple to implement, have appeared in similar forms before, and ultimately would be discoverable by any dedicated team intent on misusing LLMs.

3

107

5

2

24K

0

2

0

7K

almost 4 years ago

Useful exercise for students in #NLProc: what values are implicitly communicated through this tweet/paper? (For context, see @Abebab et al. https://t.co/3nJintYamv) #EthNLP

Ethan Mollick

@emollick

almost 4 years ago

We are running out of a vital resource: words! There are “only” 5 to 10 trillion high-quality words (papers, books, code) on the internet. Our AI models will have used all of that for training by 2026. Low-quality data (tweets, fanfic) will last to 2040. https://t.co/hm1EaJ6Enu

emollick's tweet photo. We are running out of a vital resource: words!

There are “only” 5 to 10 trillion high-quality words (papers, books, code) on the internet. Our AI models will have used all of that for training by 2026. Low-quality data (tweets, fanfic) will last to 2040. https://t.co/hm1EaJ6Enu https://t.co/PNsyUbPhBZ

68

2K

271

415

600K

2

50

6

21

17K

Daniela Tafani @DanielaTafani

almost 4 years ago

Apropos of the complete lack of transparency about #ChatGPT 's training data, I'd like to resurface what Batya Friedman and I wrote in 2018 in the original Data Statements paper: Source https://t.co/HvR7XXhDpO #NLProc #ethNLP #AIethics

emilymbender's tweet photo. Apropos of the complete lack of transparency about #ChatGPT 's training data, I'd like to resurface what Batya Friedman and I wrote in 2018 in the original Data Statements paper:

Source https://t.co/HvR7XXhDpO

#NLProc #ethNLP #AIethics https://t.co/sa02FPcPe5

0

50

17

11

0

almost 4 years ago

"I need something refreshing against your alarmism" is perfect for @emilymbender #AIEthics or #ethNLP bingo card https://t.co/09uUKc382L

Jack Stilgoe @Jackstilgoe

almost 4 years ago

Imagine thinking that books with 'menacing titles such as “Weapons of Math Destruction” and “Algorithms of Oppression”' are the thing that needs to be counterbalanced. Wow, @TheEconomist https://t.co/DaHZ1OO0Xb

3

45

3

8

0

2

1

0

over 4 years ago

Relevant for #ethnlp/#ethicalCS people: servers don't just consume electricity.

0

over 4 years ago

Interesting for #ethnlp people:

Divyansh Kaushik

@dkaushik96

over 4 years ago

Preprint alert 🚨 With ML’s growing reliance on crowdsourcing, in this paper, @zacharylipton, @AlexJohnLondon, and I seek to resolve the human subject status of ML’s crowdworkers. More in the thread🧵 1/15 https://t.co/iBuggj4OaB

2

162

35

56

0

2

0

1

0

over 4 years ago

As an #ethNLP case study, this one is interesting, because there is less (but not no) automation between the data and the insufficiently contextualized interpretation of it. So I think it might make a good model to understand what's going on with the application of lg models. >>

1

9

0

over 4 years ago

Really interesting case study to think about from the perspective of #ethNLP -- and also what linguists can contribute to society. >> https://t.co/sVVDOGeTR4

5

96

49

40

0

over 4 years ago

Ready for discussions of #ethNLP and ethics review at NLP/ML etc conferences? Don't forget your bingo card! (With @KarnFort1 on the TGV from Perpignan to Paris).

emilymbender's tweet photo. Ready for discussions of #ethNLP and ethics review at NLP/ML etc conferences? Don't forget your bingo card! (With @KarnFort1 on the TGV from Perpignan to Paris). https://t.co/qMfHNuzwpv

11

555

144

79

0

over 4 years ago

I’d love it if people share their own “committed” examples here below. I know a few more myself, but I think the more such examples the better the chances that the message (and this critical way of thinking) gets across! #AIEthics #ethnlp #ethicalCS #EthicalAI #ethics

1

2

0

Alan Ramponi @alanramponi

over 4 years ago

I like this question. Baffled by all the outrage around AI ethics? Would like to form your own opinion but not sure how to start thinking about it? Here’s my 3 minute tweetorial / lay of the land: #ethnlp #AIEthics #EthicalAI

Yoel Inbar @yorl

over 4 years ago

I'd be very interested in the strong argument against this kind of research (ideally including readings). I see a lot of moral outrage in the replies/quote tweets but I would like to know more about why. As an outsider it seems sort of baffling.

13

20

2

5

0

2

4

2

4

0

over 4 years ago

Happy to have this work accepted at #naacl2022 main conf w/ @sara_hlt! 🥂 We disentangle lexical artifacts in hate speech detection, study their impact on OOD fairness and robustness, and propose baselines and artifact statements Preprint/code available soon! #NLProc #EthNLP 🚀

DH Group at FBK @DH_FBK

over 4 years ago

New paper accepted at #NAACL2022! 🍾 "Features or Spurious Artifacts? Data-centric Baselines for Fair and Robust Hate Speech Detection" by @alanramponi and @sara_hlt. Abstract here: https://t.co/2K5QlDnnt8

0

8

1

0

5

0

Ali Alkhatib @_alialkhatib

over 4 years ago

See thread up & down, include QTs. There's a real need to get info out about how ML works, and what the privacy implications are of letting your data become part of the training data. #ethNLP #AItethics

over 4 years ago

fuck it, i'll sketch out the scenario article starts w/Uyghur activist, so let's use that. does his name transcribe correctly? probably not initially, but the journalist will fix it, retraining the model. attacker records/uploads names of Uyghur ppl; flag correctly-spelled names

1

8

2

1

0

1

12

3

1

0

Debora Nozza @debora_nozza

almost 5 years ago

Have you heard about data statements and thought it might be a good idea to write one for your #NLProc project ... but not known where to start? We have resources for you here! https://t.co/dlOMS4iyye w/@mcmillan_majora and Batya Friedman #ethNLP

2

28

11

7

0

almost 5 years ago

Can't wait for the @NLPEthics reading group to start! Check out this cool initiative by @CurriedAmanda and @anna_wegmann and fill the form ⤵️ #NLProc #EthNLP

NLPEthics @NLPEthics

almost 5 years ago

Should we really do this? What if there are unforeseen negative effects of this research question? Are there enough positive effects? The @NLPEthics reading group will start soon! We will discuss #ethics perspectives in the field of #NLP. Show interest: https://t.co/RRj2EBBBww

0

12

7

2

0

5

0

almost 5 years ago

Got opinions about ACL ethics review? Feel like there should be some community input into that process? Here's the ACL Ethics committee's community survey. Please take a few minutes and fill it out! #NLProc #ethnlp #AIEthics

0

13

9

0

almost 5 years ago

Dear #NLProc/#facct/#ethnlp people, do you know of any *qualitative* studies of either system quality or impact of a particular technology? E.g. interviews, focus groups, case studies, …

13

18

6

5

0