Voozh

Yejin Choi

2,000 posts

👁 user avatar

Yejin Choi

@YejinChoinka

professor at Stanford, researcher at NVIDIA, adventurer at heart

Seattle, WA

Joined August 2017

Pinned
👁 user avatar
Yejin Choi
@YejinChoinka
Jan 25
Excited to share TTT-Discover (Test-Time Training for Discovery)—seeking new discoveries on long-standing problems: ✅Erdős min overlap, ✅denoising for single-cell analysis, and ✅GPU kernels! The key insight: Scientific discovery requires learning from a long sequence of
You’re unable to view this Post because this account owner limits who can view their Posts. Learn more
👁 user avatar
Yejin Choi
@YejinChoinka
May 24, 2022
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
👁 Image
👁 user avatar
Yejin Choi
@YejinChoinka
Aug 29, 2025
Honored to be back on TIME100 AI for 2025 — alongside my longtime heroes @drfeifei and @BarzilayRegina! 😍 The recognition goes to my amazing students and colleagues, who strive to find ways to use AI to better humanity, as opposed to making AI for the sake of making AI better
👁 user avatar
Yejin Choi
@YejinChoinka
Oct 9, 2020
I just had a revelation that of 11 papers at @emnlp2020 (including Findings) for which I am honored to be a co-author, 🌈7 papers🌈 had 😎a woman😎 as the first author 😍 @Lianhuiq @anmarasovic @swabhz @VeredShwartz @hjrashkin @rachelrudinger Xinyao @uwnlp @allen_ai @ai2_mosaic
👁 user avatar
Yejin Choi
@YejinChoinka
Feb 1, 2019
Can neural networks learn commonsense reasoning about everyday events? Yes, if trained on 🔥ATOMIC: an Atlas of Machine Commonsense🔥, a graph of 870,000 if-then relations in natural language. @MaartenSap at #AAAI19 today 11:15am. homes.cs.washington.edu/~msap/atomic/@nlpnoah @uwnlp @allen_ai
👁 Image
👁 user avatar
Yejin Choi
@YejinChoinka
Aug 17, 2018
We have a hot new dataset for 🔥NLI with commonsense🔥! --- 113k multiple choice questions adversarially constructed to make your favorite LM/NLI/QA models struggle 😎 Stay tuned for the leaderboard going online in two weeks! #nlproc #emnlp2018 @uwnlp
👁 user avatar
Rowan Zellers
👁 Thinking Machines
@rown
Aug 17, 2018
Announcing SWAG, a new natural language inference dataset, to appear at #emnlp2018. We present a general framework for collecting adversarial QA pairs at scale, minimizing bias. With @ybisk, @royschwartz02, @YejinChoinka. rowanzellers.com/swag/arxiv.org/abs/1808.05326
👁 Two examples from SWAG, our new dataset.
👁 user avatar
Yejin Choi
@YejinChoinka
Oct 13, 2025
Instruction tuning has a hidden cost: ✅ Better at following instructions ❌ Narrower output distribution ❌ Worse in-context steerability We built 🌈 Spectrum Suite to investigate this and 🌈 Spectrum Tuning as an alternative post-training method —
👁 user avatar
Taylor Sorensen
@ma_tay_
Oct 13, 2025
🤖➡️📉 Post-training made LLMs better at chat and reasoning—but worse at distributional alignment, diversity, and sometimes even steering(!) We measure this with our new resource (Spectrum Suite) and introduce Spectrum Tuning (method) to bring them back into our models! 🌈 1/🧵
👁 Image
👁 user avatar
Yejin Choi
@YejinChoinka
Nov 3, 2019
The⏱TimeTravel⏱dataset of our #EMNLP2019 paper, 🎞Counterfactual Story Reasoning and Generation🎞 (arxiv.org/abs/1909.04076) tests counterfactual reasoning over events that unfold over time, directly addressing @GaryMarcus's call for a challenge against current neural models....
👁 Image
👁 user avatar
Gary Marcus
@GaryMarcus
Oct 29, 2019
Working on a benchmark to test the capacity of contemporary AI to develop anticipate how events unfold over time, given natural language input. If anyone wants to - help make the benchmark or - compete let me know (contact form at garymarcus dot com)
👁 user avatar
Yejin Choi
@YejinChoinka
Dec 5, 2020
Replying to @_KarenHao @timnitGebru and @emilymbender
Oh I realize I've seen their working manuscript couple weeks ago (as I was asked for feedback) and i thought it was a fantastic paper with thoughtful, well-argued discussions and extremely intensive related work...
👁 user avatar
Yejin Choi
@YejinChoinka
Feb 9, 2020
an incredibly lucid, witty, and insightful talk 😍 by @ybisk
👁 Image
👁 user avatar
Anna Rogers
@annargrs
Feb 9, 2020
Yonathan Bisk at @RealAAAI: PIQA: Reasoning about Physical Commonsense in Natural Language New benchmark for reasoning about ways to do things, targeting commonsense knowledge Paper: arxiv.org/pdf/1911.11641…@ybisk @rown @YejinChoinka
👁 user avatar
Yejin Choi
@YejinChoinka
Apr 25, 2022
Keynote by @_beenkim on Interpretability & Alignment at #ICLR2022 is live now! Come ask questions through RocketChat at iclr.cc/virtual/2022/i… Also check out the book "The Alignment Problem" (my recent favorite) by @brianchristian featuring her work and her inspiring backstory!
👁 Image
👁 user avatar
Yejin Choi
@YejinChoinka
Jul 10, 2020
It felt real special to participate in @FEVERworkshop and learn about amazing recent progress, e.g., @IAugenstein's keynote addressing various subcomponents of the whole pipeline of fact-checking, including "generating fact-checking explanations" (arxiv.org/abs/2004.05773) 1/N
👁 Image
👁 user avatar
Isabelle Augenstein
@IAugenstein
Jul 9, 2020
Don't forget that #acl2020nlp @FEVERworkshop on #factchecking is happening today from 11 GMT / 4 PDT fever.ai/workshop.html virtual.acl2020.org/workshop_W5.ht… Invited speakers: @noamslonim, myself, @roozenbot, @psresnik, Dilek Hakkani-Tür, @YejinChoinka Looking forward to it 😀 #NLProc
👁 user avatar
Yejin Choi
@YejinChoinka
Nov 3, 2019
😍Social IQA😭 (arxiv.org/abs/1904.09728), the EQ test for AI at #emnlp2019, trivial for humans, hard for neural models, the kind @GaryMarcus wants more of. It's also a resource for transfer learning of commonsense knowledge, achieving new SOTA on related tasks (Winograd, COPA).
👁 user avatar
Yejin Choi
@YejinChoinka
May 31, 2023
Importantly, we distill the student model from the ever-so-shabby GPT-2 as the teacher model, instead of distilling from significantly more powerful/larger LLMs such as GPT-3, thus the name, 🔥impossible distillation🔥
👁 user avatar
Peter West
@PeterWestTM
May 30, 2023
If you're overloaded hearing about large/expensive/proprietary models check out our preprint: Impossible Distillation‼️ smaller, off-the-shelf LMs (e.g. GPT-2) still have something to offer -- by generating high-quality task data, even for tasks they can't directly solve

URL: https://x.com/YejinChoinka

⇱ Yejin Choi (@YejinChoinka) / X