Joined June 2022
- Tweets42
- Following270
- Followers154
- Likes175
Connor Baumler retweeted
🚨 New Paper! 🚨
Your agent is writing code faster than you can understand it 😬
In our controlled online study, coding agents helped users finish tasks quickly, but reduced how well they understood the code they just “wrote”
We also find three ways to avoid this tradeoff 👇🧵
Connor Baumler retweeted
Excited to share two papers at #EMNLP2026! 🥳
1️⃣ Check the Scoreboard: We extend *scoring schemes* from education research and validate their benefits for LLM evals (Main) 🧑🏫📈
2️⃣ VibeJam: Our system for human-AI studies with coding agents (Demo) 🧑💻🌐
See you in Budapest! 🤠
ALT emnlp :)
Connor Baumler retweeted
Can you edit AI writing to sound like you?
In collaboration with Pangram, researchers at the University of Maryland found that even after editing, LLM text remained more similar to AI writing than to human — despite editor's reports that the edited text reflected their voice.
Connor Baumler retweeted
Ever wonder how LLMs perceive gender from first names (and occupations)? 🤔
🔗 arxiv.org/abs/2503.06792
Our paper (w/ @ctbaumler, @AbhilashaSanch2, @rachelrudinger) finds that:
(1) LLMs representations of gender for first names mirror real-world gender distributions. 📊
Connor Baumler retweeted
Automating AI research is exciting! But can LLMs actually produce novel, expert-level research ideas?
After a year-long study, we obtained the first statistically significant conclusion: LLM-generated ideas are more novel than ideas written by expert human researchers.
Connor Baumler retweeted
A cool graphic from @haldaume3 @genderbiasnlp keynote: the impact of text completion on stereotypes stories. Pro stereotyped suggestions are the same or worse than no text completion, but anti stereotype can reduce it. #ACL2024
Connor Baumler retweeted
How can we humans verify the truthfulness of LLM outputs (or any claims you see on the Internet)? Should we ask ChatGPT (#LLMs)? Search on Google (retrieval)? Are they complementary?
Tldr: LLMs Help Humans Verify Truthfulness - Except When They Are Convincingly Wrong!
1/n
Connor Baumler retweeted
Does Writing with Language Models Reduce Content Diversity?
TL;DR: Yes! But it depends on which language model you use 🤖🕵️
Sharing work with my advisor @hhexiy: arxiv.org/abs/2309.05196
Code/data: github.com/vishakhpk/hai-div…
#NLProc #paper #LLMs
Connor Baumler retweeted
This talk is starting soon (in ~30 min). See you all there!!
🧭📍In-person participants, come to Pier 2&3 (on first floor right above the venue lobby)
👩💻🧑💻Virtual participants, find the zoom link here: underline.io/events/395/sess…
LLMs say stuff.
Some is factual, some not.
Some is contextually appropriate, some not.
In both cases, Grice is unhappy.
Discerning these two is important, especially when people struggle to detect factuality errors. In our #ACL2023 paper, we give evaluation strategies to do so.
Connor Baumler retweeted
We develop a framework for measuring factual errors—facts about entities that are nonfactual, but seem contextually plausible, and contextual errors—facts about entities that are factual, but contextually irrelevant, in referring expression generation setting.
Connor Baumler retweeted
Can we better capture harms in text generation? Our #ACL2023 paper introduces FairPrism, a dataset with detailed human annotations on AI-generated text to diagnose fairness-related harms caused by text generation systems & flaws of mitigation tools🧵
📄bit.ly/fairprism
Connor Baumler retweeted
We are getting ready to start our workshop at #ACL2023NLP today! We have an exciting program with 8 paper presentations, a panel on Collective Action and two wonderful keynotes by @lexicondk and @juanmvsa. Come meet us at Pier 9 or online at 9:00 am! queerinai.com/acl-2023
Connor Baumler retweeted
Queer in AI is excited to announce a Call for Contributions for our affinity workshop at @aclmeeting ! More details in the thread below. #NLProc
TL;DR: we have two deadlines: a visa-friendly one (May 12) and a final one (June 9). Details on our website: queerinai.com/acl-2023
Connor Baumler retweeted
When NLI models can ignore context, do they? Our #NAACL2022 paper explores this + revisits the role of partial-input baselines in NLI! Joint work w/ @rachelrudinger
🔗 arxiv.org/abs/2205.12181
I'll be presenting this today:
📆 In-Person PS 3 (10:45 PT)
📆 Virtual PS 2 (14:15 PT)
ALT Given NLI instances that partial-input baselines correctly label, we edit their context to induce a different gold label.
Connor Baumler retweeted
Researchers @ClipUmd are presenting a plethora of papers and workshops on human-centered natural language processing this week @naaclmeeting.
Read more: go.umd.edu/cqP #NAACL2022
