🤖 ACL 2021 paper 🤖 We introduce DExperts, a decoding-time method for controlled text generation which combines a pretrained LM with experts 😇 and/or anti-experts 😈 in an ensemble of LMs. 🧵 📄: arxiv.org/abs/2105.03023 💻: github.com/alisawuffles/D…

"WinoGrande," by Keisuke Sakaguchi, Ronan Le Bras, and Chandra Bhagavatula Ai2, and Yejin Choi Allen School, introduces the WinoGrande dataset with 44k problems inspired by the design of the WSC, but modified to improve the scale and hardness of the problems. bit.ly/3gzYZNs

Introduce our new preprint—Delphi: Towards Machine Ethics and Norms arxiv.org/abs/2110.07574 ✨Delphi is a commonsense moral model with a robust performance of language-based moral reasoning on complicated everyday situations. ✨Ask Delphi demo at: delphi.allenai.org (1/N)

HUGE thanks to my many creative and talented coauthors: Chandra Sekhar, Jack Hessel, Jena D. Hwang, Liwei Jiang , Ronan Le Bras , Ximing Lu, Sean Welleck and Yejin Choi And thank you to AK moved to @_akhaliq for sharing this before I even woke up (2/2)

This is a joint work with many amazing collaborators Ai2 and Allen School: Jena D. Hwang, Chandra Sekhar, Ronan Le Bras, Max Forbes, Jon Borchardt, @_jennyliang, Oren Etzioni, Maarten Sap (he/him), Yejin Choi Welcome to check out our ✨Ask Delphi✨ demo at: delphi.allenai.org

We also present a list of important open questions and avenues for future research in our preprint. We sincerely urge our research community to collectively tackle these research challenges head-on, in an attempt to build ethical, reliable, and inclusive AI systems.

Really nice work on machine ethics from Ai2, indeed. So nice to see that commonsense reasoning boosts machine ethics. Looking forwards to boost our work of Patrick Schramowski (arxiv.org/abs/2103.11790) via commonsense and actually some explicit rules. Towards hybrid machine ethics!

Very interesting demo, I tried many examples and got the answers expected. But, it looks like #Delphi has developed its own opinions about itself 😁 Yejin Choi

So the CS/AI/NLP folks Ai2 have put up a "computational model for descriptive ethics" call #delphi that's fascinating both for what it does and doesn't do...

Our team Ai2 is looking for interns for summer 2022! Consider applying:

Great coverage of the potential and the limitations our researchers are grappling with in the new field of machine ethics, by Charles Q. Choi 🚀 in IEEE Spectrum: spectrum.ieee.org/ai-ethics-mach…

Today on the AI2 Blog: An overview of our model Delphi as a nascent step toward machine ethics, a look into the stress testing of this research prototype, the lessons learned, and reflection on why working on imperfect solutions is better than inaction: medium.com/ai2-blog/towar…

I am extremely honored and excited to give a keynote at the 60th ACL today 🚀2082: An ACL Odyssey: The Dark Matter of Intelligence and Language🚀in which I will share reflections on the past ACL and weird speculations on the future, thus the retro-futuristic theme #acl2022nlp 1/N

Deadline to apply for a summer 2023 internship on the Mosaic team is coming up soon (November 13)!

⚡️New preprint! 📰 We introduce a question generation approach for disambiguating social and moral situations using RL. With: Liwei Jiang, Ximing Lu, Vivek Srikumar, Jena Hwang, Chandra Sekhar, Yejin Choi Preprint: arxiv.org/pdf/2212.10409…

ꜱᴄᴀʟᴇ ɪꜱ ɴᴏᴛ ᴛʜᴇ ᴏɴʟʏ ᴡᴀʏ! Check out our #ACL2023 poster on I2D2 at 9AM in the Frontec Ballroom.

Check out our #ACL2023 poster on I2D2 at 9AM in the Frontec Ballroom.

We're thrilled to introduce Satlas, a brand new platform from AI2 for exploring global geospatial data generated by AI from satellite imagery. 🛰️ Try it out: satlas.allen.ai Learn more: blog.allenai.org/satlas-monitor…

Human values are crucial to human decision-making. We may consider health or environmental responsibility when deciding to bike to work, or weigh loyalty against honesty when deciding if we should lie to protect a friend. (0/n)

"Less (tuning) is more for alignment" is an intriguing hypothesis. Is alignment tuning really that “superficial”⁉️ 🤔 If so, how so? 🤔 Can any straightforward analysis explain this? 🤔 What if I tell you “no tuning can also be great for alignment”? 🫢 😉 If you’re interested in