Peter West (@PeterWestTM) / X

Peter West

281 posts

Peter West

@PeterWestTM

AI / NLP Researcher Incoming faculty at @UBC_CS and @CAIDA_UBC Postdoctoral fellow at @StanfordHAI @stanfordnlp Former PhD student at @uwcse @uwnlp he/him

peterwest.pw

Joined September 2019

Pinned
Peter West
@PeterWestTM
Dec 3, 2024
I have multiple MSc/PhD openings in my lab at @UBC_CS! Come discover the hidden capabilities/limits of LLMs, e.g. how to learn from, guide, and understand the outputs of models. See my website (bio) for more details. cs.ubc.ca/students/grad/… Apply by December 15th! Also...
23K
Peter West
@PeterWestTM
Nov 3, 2023
Richard Feynman said “What I cannot create, I do not understand”💡 Generative Models CAN create, i.e. generate, but do they understand? Our 📣new work📣 finds that the answer might unintuitively be NO🚫 We call this the 💥Generative AI Paradox💥 paper: arxiv.org/abs/2311.00059
156K
Peter West
@PeterWestTM
Aug 23, 2024
I'm very excited to be starting my dream job as faculty at @UBC_CS @CAIDA_UBC in 2025 and postdoc-ing with @ChrisGPotts at @StanfordHAI @stanfordnlp this year! I am recruiting students this cycle who are curious to explore the mysteries and limitations of LMs / GenAI ...
37K
Peter West
@PeterWestTM
Oct 15, 2021
I'm excited to introduce Symbolic Knowledge Distillation! ⭐️Commonsense knowledge with a twist: generated by models 🤖 filtered by models 🤖 & learned by models 🤖 ⭐️Human-quality knowledge, with (almost) no humans 🙅 ⭐️A student model 🤖 that beats its teacher 🧑‍🏫 and more! (1/2)
Peter West
@PeterWestTM
Mar 23, 2022
Factuality is a key concern in NLG, yet most attention goes to summarization. I’m excited to announce Factual Ablation, a new eval for the grounded content transfer task to study, measure, and improve factuality paper: arxiv.org/abs/2203.10133 code: github.com/peterwestuw/Fa… 1/2
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
AK
@_akhaliq
May 29, 2023
Impossible Distillation: from Low-Quality Model to High-Quality Dataset & Model for Summarization and Paraphrasing propose that language models can learn to summarize and paraphrase sentences, with none of these 3 factors. We present Impossible Distillation, a framework that
22K
Peter West
@PeterWestTM
Sep 24, 2019
Sentence summarization with Information Bottleneck 🍾 and no supervision! 😎 Self-supervised and unsupervised approaches 🍾BottleSum🍾 to appear at @emnlp2019 arxiv.org/abs/1909.07405 With @universeinanegg @janmbuys @YejinChoinka at @uwcse and @allen_ai
arxiv.org
BottleSum: Unsupervised and Self-supervised Sentence Summarization...
The principle of the Information Bottleneck (Tishby et al. 1999) is to produce a summary of information X optimized to predict some other relevant information Y. In this paper, we propose a novel...
Peter West
@PeterWestTM
Jul 20, 2022
We're updating our commonsense knowledge graph (ATOMIC-10X) and model (COMET-distill) to include 6 more commonsense relations! v2.0 of our data and models here github.com/peterwestai2/s… What commonsense relations would you like to see as LMs make this process easier and easier?
Peter West
@PeterWestTM
Oct 15, 2021
I'm excited to introduce Symbolic Knowledge Distillation! ⭐️Commonsense knowledge with a twist: generated by models 🤖 filtered by models 🤖 & learned by models 🤖 ⭐️Human-quality knowledge, with (almost) no humans 🙅 ⭐️A student model 🤖 that beats its teacher 🧑‍🏫 and more! (1/2)
Peter West
@PeterWestTM
Jun 17, 2022
Honored that we received a best paper award at NAACL and to be a co-2nd author on it! If you haven't, I suggest checking out @GXiming's body of work--filled with similarly creative, high-quality, and unique research. Truly a rising star in NLP!
Sean Welleck
@wellecks
Jun 17, 2022
Honored to receive a Best Paper award at NAACL 2022 for NeuroLogic A*esque Decoding, with an awesome team @GXiming @PeterWestTM @liweijianglw @wittgen_ball @DanielKhashabi @Ronan_LeBras @Lianhuiq @YoungjaeYu3 @rown @nlpnoah @YejinChoinka ! arxiv.org/abs/2112.08726
Peter West
@PeterWestTM
Jul 13, 2022
Very excited to present Symbolic Knowledge Distillation at NAACL tomorrow (July 13)! For those early risers, I will be speaking at the ML for NLP session (8:00 - 9:00 a.m) at 8:30 in Quinault. For those who sleep in, come find me for a chat around the conference!
Peter West
@PeterWestTM
Oct 15, 2021
I'm excited to introduce Symbolic Knowledge Distillation! ⭐️Commonsense knowledge with a twist: generated by models 🤖 filtered by models 🤖 & learned by models 🤖 ⭐️Human-quality knowledge, with (almost) no humans 🙅 ⭐️A student model 🤖 that beats its teacher 🧑‍🏫 and more! (1/2)
Peter West
@PeterWestTM
Jul 30, 2021
Excited to share Reflective Decoding at ACL 2021! 🤖📣↔️ Take LM decoding beyond unidirectional by decoding twice: just generate the context ➡️ then go back ⬅️ to paraphrase & infill Come chat at poster session 1Q! Code: reflectivedecoding.com Paper: arxiv.org/pdf/2010.08566…
Peter West
@PeterWestTM
Jul 14, 2022
A late stage comeback for the donut wall at NAACL2020
Peter West
@PeterWestTM
Nov 3, 2023
Replying to @PeterWestTM
Just because a model can generate an answer to a question, this doesn’t mean it understands well enough to answer discriminatively, between multiple choices. However, Humans behave more intuitively: their ability to generate tends to come with the ability to discriminate too
4.8K
Peter West
@PeterWestTM
Aug 23, 2024
Replying to @PeterWestTM
Finally, I want to thank my family, friends, and mentors for seeing me to the end of this PhD and onto my dream job. There are far too many people to name here, evidence that every great achievement is shared Especially I thank @YejinChoinka for her wisdom, creativity, and care.
1.3K