Ryan He

[navj @pranav_jad]

RLMs give you long context so why not share chat histories across a repo. Thanks @modal for the great platform and @hackwithtrees for a great event!

[Michael Siu @michaelsiux]

We built Monolith, an RLM-as-a-service MCP tool that won at TreeHacks 2026!

AI coding agents lose all context between sessions. We solve this by using Recursive Language Models to let agents reason over arbitrarily large contexts with persistent memory across sessions.

[Michael Saxon @m2saxon]

🚨We've been assessing T2I metrics wrong...until now‼️

Our new meta-metric for T2I faithfulness metrics, T2IScoreScore (TS2), checks if a metric correctly orders and separates many images against single prompts!

The results may surprise you...

https://t.co/Afr61aiL9o

1/5

[William Wang @WilliamWangNLP]

One thing I'd like to have at EMNLP/*ACL is an EARLY CAREER SPOTLIGHT session that gives opportunities for new faculty, research scientists, postdocs, and sr. grad students to talk about BLUE SKY IDEAs. The field needs more discussions about the big picture and the road forward.

[William Wang @WilliamWangNLP]

Sharing my vision for SIGDAT:
1️⃣ Advocate Open Science Research
2️⃣ Ensure consistent review practices & Allow finding papers be presented at EMNLP.
3️⃣ Commit to aggressive fundraising for ECRs & champion an evidence-based approach to inclusion.

[Wenda Xu @WendaXu2]

[New paper!] Can LLMs truly evaluate their own output? Can self-refine/self-reward improve LLMs? Our study reveals that LLMs exhibit biases towards their output. This self-bias gets amplified during self-refine/self-reward, leading to a negative impact on performance. @ucsbNLP

[Liangming Pan @PanLiangming]

📄 Check out our new paper on analyzing the self-correction capability of large language models. 🔥

[AerIn @aerinykim]

One thing that stood out to me at #NeurIPS2023 was there were a lot more application papers than before.
This was not the case with my last neurips (2018).

I'd like to encourage this wave by highlighting one such paper.

LayoutGPT: Compositional Visual Planning and Generation with Large Language Models
https://t.co/CBSFjvbNsU

1. What it does

With a text prompt, for example, "A living room with a sofa, a coffee table, ...", the app generates an image as well as the layout of the objects - their absolute position in pixel coordinates. e.g. {height: 81px; width: 93px; top: 119px; left: 15px;}

2. Why it's useful

You might wonder how this is different from stable diffusion.

Stable Diffusion: Text -> Image.
LayoutGPT: Text -> Layout -> Image.

Adding the layout step and before using generative models gives you control over where things are placed. The author suggests many practical applications such as text-guided dense layout generation, inpainting, scene synthesis/completion, etc.

In my experience, marking human pose keypoints is quite difficult, yet this model does a pretty good job at it.

They even showed how their method can be extended to 3d spaces.

3. How it's done

Doing something new?
This means you need a new dataset.

What they did was take bounding boxes from MS COCO and use them to create the object-coordinate dataset. When the user inputs the prompt, it searches similar images and labels on the fly and adds them in-context.

[William Wang @WilliamWangNLP]

Liangming Pan is presenting Logic-LM at #emnlp2023 right now. #nlproc

[William Wang @WilliamWangNLP]

UCSB #NLProc and alumni at #EMNLP2023! 🇸🇬🇸🇬🇸🇬🥳🥳🥳

[William Wang @WilliamWangNLP]

Our undergraduate students present video reasoning work VIP at #EMNLP2023.

[Deepak Nathani @deepaknathani11]

Presenting tomorrow (Dec 08, 14:00) at #EMNLP2023

MAF: Multi-Aspect Feedback for Improving Reasoning in Large Language Models

Work done with my amazing collaborators and advisors: @DayvidWang, @PanLiangming, @WilliamWangNLP

[Michael Saxon @m2saxon]

Go say hi to our @ucsbNLP undergrads at #EMNLP23 and check out their dataset on assessing sequential visual reasoning in VLMs!

[Vaishnavi H @vaishnavihima]

🚀 We are happy to officially release the VIP dataset! #videoChainofThought #videoreasoning #LLMs #EMNLP2023

Paper: https://t.co/e3xYj7Y5af
Dataset: https://t.co/1DZG0bEeJq
🧵1/6

[Tony Sun @tony_sun9]

My first blog post: “An Intro to Real-Time Machine Learning: Computing Features for Prediction.”

This post is for data scientists and machine learning engineers who want to gain a better understanding of the underlying data pipelines to serve features for real-time prediction.

[Alex Mei @alexmeigz]

🚨 RED ALERT! 🚨
Tired of expensive and intensive testing of your language model? It's time to automate red teaming with our latest research project accepted to #EMNLP2023: ASSERT -- a test suite for Automated Safety ScEnario Red Teaming! https://t.co/YtDkxc5yAS

🧵1/n #NLProc

[Liangming Pan @PanLiangming]

What if LLM-generated misinformation pollutes the corpus (or even the Internet) we use for retrieving information? We studied how it affects Open-domain QA in our paper accepted to Findings of #EMNLP2023.

Paper: https://t.co/daAxldJZ7O
Code: https://t.co/W6J5Z9taLD

[Zhiyu Zoey Chen @ZhiyuChen4]

I will be joining @UT_Dallas as an Assistant Professor in @UTDJonsson in Fall 2024. Deep thanks to my advisors @WilliamWangNLP, Xifeng Yan, collaborators, and friends for supporting me as always. Currently, I'm working as a postdoc in @S3DatCMU. Look forward to the new journey!

[William Wang @WilliamWangNLP]

🎉 Thrilled for @ZhiyuChen4 joining @UT_Dallas! As her PhD advisor, I've witnessed her NLP milestones. Her 2019 Few-Shot NLG paper & FinQA contributions are transformative. UTD's NLP group is stronger with her. Students, seize the chance to learn from one of the best! 🌟 #NLProc

[Michael Saxon @m2saxon]

Our work, "Let's Think Frame by Frame with VIP: A Video Infilling and Prediction Dataset for Evaluating Video CoT" got in EMNLP23! Congrats undergrad coauthors @vaishnavihima @DannyRose30 @andy__ouyang @RyanHe88 on your 1st paper!

Stay tuned for arxiv update and dataset link!