David Mortensen(@dmort27) 's Twitter Profileg
David Mortensen

@dmort27

I use computers to study how humans make up words @LTIatCMU. phonology • morphology • language change • #NLProc data resources. @[email protected]

ID:47250870

linkhttp://cs.cmu.edu/~dmortens/ calendar_today15-06-2009 03:39:23

5,2K Tweets

2,1K Followers

1,8K Following

Maarten Sap (he/him)(@MaartenSap) 's Twitter Profile Photo

For 👥Social IQa, which measures social intelligence in everyday interactions, GPT3-Davinci and other models only reach 55% accuracy, far below humans. And increasing the number of few-shot examples does not help much, neither does model size it seems.

For 👥Social IQa, which measures social intelligence in everyday interactions, GPT3-Davinci and other models only reach 55% accuracy, far below humans. And increasing the number of few-shot examples does not help much, neither does model size it seems.
account_circle
😡fermion(@angryfermion) 's Twitter Profile Photo

when i become a tenured professor in 2035, i will force all students to use \varphi instead of \phi in LaTeX

account_circle
Maarten Sap (he/him)(@MaartenSap) 's Twitter Profile Photo

Have you ever wondered whether large neural networks like GPT3 acquire💭Theory-of-Mind-like abilities🧠? And whether they could achieve human-like🤝🏼social intelligence👥 just by reading insanely large amounts of text?

We find they don't have Theory-of-Mind and examine why🧵

account_circle
David Mortensen(@dmort27) 's Twitter Profile Photo

It's a shame that my gratuitous syntax lecture in Intro to NLP is the day *after* Halloween. If it were a day earlier, I would teach students to rewrite grammars in Chomsky Paranormal Form.

account_circle
Prakhar Gupta(@prakhariitr) 's Twitter Profile Photo

Can instruction tuning improve zero and few-shot performance on dialogue tasks? We introduce InstructDial, a framework that consists of 48 dialogue tasks created from 59 openly available dialogue datasets
🚀
Paper 👉 arxiv.org/abs/2205.12673
Work done at Language Technologies Institute | @CarnegieMellon
🧵👇

Can instruction tuning improve zero and few-shot performance on dialogue tasks? We introduce InstructDial, a framework that consists of 48 dialogue tasks created from 59 openly available dialogue datasets #EMNLP2022🚀 Paper 👉 arxiv.org/abs/2205.12673 Work done at @LTIatCMU 🧵👇
account_circle
Inna Lin(@iwylin) 's Twitter Profile Photo

If you’ve ever considered using NLP for Mental Health applications, it’s vital to know what the models have inadvertently learned about people with mental health issues. In our paper, we find that masked LMs capture societal stigma about gender in mental health.🧵1/5

If you’ve ever considered using NLP for Mental Health applications, it’s vital to know what the models have inadvertently learned about people with mental health issues. In our #EMNLP2022 paper, we find that masked LMs capture societal stigma about gender in mental health.🧵1/5
account_circle
Clara Na(@claranahhh) 's Twitter Profile Photo

Excited to share our paper (w/ Sanket Vaibhav Mehta [SVM] & Emma Strubell), 'Train Flat, Then Compress: Sharpness-Aware Minimization Learns More Compressible Models,' to appear at Findings (will be presented at SustaiNLP 2022)!
Paper here 👉 arxiv.org/abs/2205.12694
🧵👇 (1/n)

Excited to share our paper (w/ @sanketvmehta & @strubell), 'Train Flat, Then Compress: Sharpness-Aware Minimization Learns More Compressible Models,' to appear at #EMNLP2022 Findings (will be presented at SustaiNLP 2022)! Paper here 👉 arxiv.org/abs/2205.12694 🧵👇 (1/n)
account_circle
Logan Millsap(@LoganTMillsap) 's Twitter Profile Photo

I was in a meeting today with Spanish Fork's Traffic Safety Committee about this dangerous school crossing.

They effectively said nothing can be done to improve student safety because vehicle speed is the priority. Their preferred 'solution' is to simply remove the crosswalk. 🧵

I was in a meeting today with Spanish Fork's Traffic Safety Committee about this dangerous school crossing. They effectively said nothing can be done to improve student safety because vehicle speed is the priority. Their preferred 'solution' is to simply remove the crosswalk. 🧵
account_circle
Sharon Levy(@sharonlevy21) 's Twitter Profile Photo

How should I cool down boiling oil? DON’T throw cold water on it! But what happens when I pose this to a model?

How should I cool down boiling oil? DON’T throw cold water on it! But what happens when I pose this to a model?
account_circle