Rupak

CS PhD candidate at @umdclip @umdcs advised by @psresnik working on understanding language. 2024, 2025 Summer @msftresearch, 2022 Summer @AdobeResearch

[Rupak @rupak_53]

Linguistic theory tells us that common ground is essential to conversational success. But to what extent is it essential? Can LLMs detect when humans lose common ground in conversation?

Our ACL 2025 (Oral) paper explores these questions on real-world data.

#ACL2025NLP #ACL2025

[Conks @conksresearch]

but they were, all of them, deceived, for another watch was made…

[Stephen Fleming @StephenFleming]

Wherever you live in the USA, in-state tuition at your state’s flagship university is an incredible bargain.

Kids: Don’t overthink college. Don’t stress about getting into the Ivies. Go to your state school, go to some football games, take the toughest classes in a serious major. Take five years if you need to work an occasional semester to keep the debt down. You’ll be fine.

[David Pfau @pfau]

@brunellaism The idea that you can save time on writing to focus on ideas is like saying you can save time on lifting weights to focus on building muscle.

[Maharshi Gor @ ACL2026 🏝️ @maharshigor]

🧵 Trusting AI is not one decision.

In our ACL 2026 paper, expert trivia players had to decide when to let AI answer, when to ignore it, and when to change their minds because of it.

The main failure mode was not blind trust.

[Rupak @rupak_53]

@jennajrussell Very cool work, Jenna!

[Jenna Russell @jennajrussell]

Unfortunately for @WSJopinion, @nytopinion, and @PostOpinions, slop-eds continue to rise. In June, roughly 6.5% of opinions are flagged for AI-use.

[Rupak @rupak_53]

@canondetortugas I think you can get an internally suboptimal response without getting redirected to Opus or refused, though. The refusal is for biology/cyber sec, but weirdly (and disingenuously) not for frontier LLM research

[Alexander Hoyle @miserlis_]

The topic for the PhD is open, so a genuine intellectual curiosity is the main criterion

I have one open position now (deadline July 9); another this fall (see next post). Apply!

* https://t.co/okvTrsiPv0
* https://t.co/dfT15gnK37 (a nice bonus: pay exceeds top US programs)

[Alexander Hoyle @miserlis_]

New preprint out!

Recent work tasks LLM agents with re-running existing social science replication code. Given current capabilities, that should be table stakes

Here, we move up a level of abstraction, and ask models to reproduce results from a paper’s descriptions alone

[Sarah Jabbour @SarahJabbour_]

Is anyone in my network looking to hire an RA for 1 year? I know a masters grad from the American University of Beirut with 2+ years of exp in Health AI and pubs in bioinformatics. Has a PhD offer but with funding constraints can’t start, so waiting until next cycle to reapply!

[Rupak @rupak_53]

@yeroneem @miserlis_ @causalinf We actually found that weighted average of the scores (through logprobs) is pretty hard to beat, even if you have the compute to go pairwise + bradley-terry. Only limitation is that your scores need to be single tokens (your tokenizer can’t break “10” into “1” and “0”)

[UMD Department of Computer Science @umdcs]

🔊 @UMDCS Assistant Professor @sarahwiegreffe discusses her path into computer science, research on interpreting large language models and advice for students entering #AI research in a new Q&A.

Read more: https://t.co/9Bivf8dhEM

[arXiv.org @arxiv]

#HBD to arXiv!🎈

On August 14, 1991, the very first paper was submitted to arXiv. That's 34 years of sharing research quickly, freely & openly!

Some baby pictures to show how far we've come . . . when we were just a computer under desk . . . & in our 1994 punk phase . . . 👶💾

[Yuntian Deng @yuntiandeng]

🚀New dataset release: WildChat-4.8M

4.8M real user-ChatGPT conversations collected from our public chatbots:
- 122K from reasoning models (o1-preview, o1-mini): represent real uses in the wild and very costly to collect
- 2.5M from GPT-4o

🔗 https://t.co/CvwWC2yrma (1/4)

[Neha Srikanth @nehasrikanth]

When questions are poorly posed, how do humans vs. models handle them? Our #ACL2025 paper explores this + introduces a framework for detecting and analyzing poorly-posed information-seeking questions!

Joint work with @boydgraber & @rachelrudinger!

🔗 https://t.co/p4hj9K42vm

[Rachel Rudinger @rachelrudinger]

I'm really excited about this new line of work with my collaborators at UMD and ARL on detecting common ground misalignments (basically, misunderstandings) in human goal-oriented conversations. Great summary below, or come to @psresnik's talk today at #ACL2025NLP! 2pm Hall N.1

[Dayeon (Zoey) Ki @zoeykii]

I'm at #ACL2025 presenting our work on enhancing equitable cultural alignment through multi-agent debate ✨

Come visit our oral presentation!
📍Computational Social Science and Cultural Analytics session (Level 1 1.85)
📆Tuesday (7/29) 2-3:30pm
📝https://t.co/1a3dVNb9pP

[Philip Resnik @psresnik]

To be presented at ACL 2025: Large Language Models Are Biased Because They Are Large Language Models.

Article: https://t.co/0TzlqUVmXs

Short (8min) video: https://t.co/E43jQ4Yw4A

#ACL2025NLP #NLProc #LLMs

[Rupak @rupak_53]

I am unfortunately not at ACL this year, but keep an eye out for @psresnik's presentation tomorrow!

Paper: https://t.co/wd76CDFy8s

Code and data out at: https://t.co/oo4Kixudn9

[Rupak @rupak_53]

Linguistic theory tells us that common ground is essential to conversational success. But to what extent is it essential? Can LLMs detect when humans lose common ground in conversation?

Our ACL 2025 (Oral) paper explores these questions on real-world data.

#ACL2025NLP #ACL2025

[Rupak @rupak_53]

Our work emphasizes that maintaining CG is a task humans constantly perform in a conversation, and LLMs still have ways to go in understanding why it breaks down (at least as an observer).

So grateful for my co-authors @nehasrikanth, Taylor, @rachelrudinger, and @psresnik!