🗞️ Pick of the week @fbk-mt.bsky.social: Most LLM dev centers current adopters, so what are we missing? Worth a read!
👇https://arxiv.org/pdf/2510.15951
#NLP #LLMs #HCI
Our #PickOfTheWeek by @bsavoldi.bsky.social: "Attention to Non-Adopters" by @kaitlynzhou.bsky.social, @gligoric.bsky.social, @myra.bsky.social, @mlam.bsky.social, @vyoma-raman.bsky.social, Boluwatife Aminu, Caeley Woo, Michael Brockman, @hannah-cha.bsky.social, @jurafsky.bsky.social (2025).
Paper title!
"Sorry, I Didn't Catch That": How Speech Models Miss What Matters Most
Work @togetherai.bsky.social with Federico Bianchi, @mbartelds.bsky.social @jameszou.bsky.social
Paper: arxiv.org/abs/2602.12249
We release two public datasets for evaluation: huggingface.co/datasets/kzh...
Workflow: (1) Select a sample of speech from Common Voice, e.g., Spanish (2) Set the XTTS to generate speech in Spanish (supports 16 languages, excluding English) (3) Clone the voice and generate Spanish but with injected English street names, e.g., "Estoy en ...Washington" (4) Extract street name speech and manually validate. Repeat this with as many samples as needed to create a unique finetuning dataset.
To mitigate this harm, we introduce a synthetic data generation approach that produces diverse pronunciations. With fewer than 1,000 synthetic samples, we can improve street name transcription accuracy by nearly 60% (relative to base models).
We find that speech models systematically struggle to transcribe named entities like street names, leading to routing errors for all users. But this impact is roughly twice as large for non-English primary speakers as for English primary speakers.
As speech models are being deployed in real-world taxi and emergency service settings, the failure to accurately transcribe named entities can cause delays and errors in critical settings.
🚀 Apply to CMU LTI’s Summer 2026 “Language Technology for All” internship! 🎓 Open to pre‑doctoral students new to language tech (non‑CS backgrounds welcome). 🔬 12–14 weeks in‑person in Pittsburgh — travel + stipend paid. 💸 Deadline: Feb 20, 11:59pm ET. Apply → forms.gle/cUu8g6wb27Hs...
Come work with Federico Bianchi, Yongchan Kwon, Shang Zhu, James Zou, and me on research at the intersection of agent capabilities, alignment, and evaluation.
📍 San Francisco (in-person)
⏱️ 12 weeks, Summer 2026
Link: job-boards.greenhouse.io/togetherai/j...
Internship opportunity! Please share!
📣 I'm looking to hire an intern in human-centered NLP for the agents team at Together AI. Come work on frontier AI systems that tackle complex agentic tasks!
Research direction is open and looking to publish in NLP and HCI venues!
Please share widely!
My research focuses on human-centered NLP, both in evaluating and training LLMs as well as designing safe and reliable human-LM interactions. More information here!
katezhou.github.io
Application fee waivers can be requested here: gradschool.cornell.edu/admissions/a...
Photo of Cornelll University building surrounded by colorful trees
No better time to start learning about that #AI thing everyone's talking about...
📢 I'm recruiting PhD students in Computer Science or Information Science @cornellbowers.bsky.social!
If you're interested, apply to either department (yes, either program!) and list me as a potential advisor!
and @hannah-cha.bsky.social
Together, we’re excited to continue work on developing LLMs for the needs of a broader user audience! This work is done in collaboration with:
@gligoric.bsky.social @myra.bsky.social @mlam.bsky.social @jurafsky.bsky.social
@stanfordnlp.bsky.social @togetherai.bsky.social
We outline several strategies known in the HCI literature to elevate non-adopter needs and integrate them into LLM development:
1️⃣ re-balancing data annotation and interaction logs
2️⃣ participatory design for developing evaluations
3️⃣ non-adopter-centered task ideation
Many non-adopters have legitimate reasons to resist adopting or to stop using chat models; however, as research practitioners who have the power to design and shape model capabilities, we urge the community to develop technologies where non-use is a choice, rather than an inevitable circumstance.
From users interviews (n=23) and an online survey (n=230), we found:
1️⃣ Non-adopters are interested in chat models, but face barriers to adoption
2️⃣ Non-adopters prioritize tasks rarely reflected in model evals: navigating healthcare portals, coordinating caregiving, contextualized IR
Many current methods rely on logs, preferences, and feedback from existing users—who represent a narrow slice of the population.
Adopter-centered methods risk widening the divide between adopters and non-adopters as datasets, benchmarks, and evaluations evolve around current adopter needs.
A circular flow diagram that compares current and proposed practices for LLM development using data from adopters and non-adopters. Three gray boxes represent current practices: “R&D,” “Chat Models,” and “Adopters’ Needs and Usage Data,” connected in a clockwise loop with black arrows. A blue box labeled “Non-adopters’ Needs and Usage Data” adds a proposed feedback path, shown with blue arrows, linking non-adopter data back to R&D and adopters’ data.
As of June 2025, 66% of Americans have never used ChatGPT.
Our new position paper, Attention to Non-Adopters, explores why this matters: AI research is being shaped around adopters—leaving non-adopters’ needs, and key LLM research opportunities, behind.
arxiv.org/abs/2510.15951
We are excited to welcome five new faculty members to Cornell Bowers this semester!
A Big Red Bowers Welcome to Sasha Golovnev, Andrew Owens '10, David Rand '04, Benjamin Shestakofsky, and Kaitlyn Zhou. 👋
Read more: lnkd.in/ewA4U4fU
Screenshot of paper title: Sycophantic AI Decreases Prosocial Intentions and Promotes Dependence
AI always calling your ideas “fantastic” can feel inauthentic, but what are sycophancy’s deeper harms? We find that in the common use case of seeking AI advice on interpersonal situations—specifically conflicts—sycophancy makes people feel more right & less willing to apologize.
It is PhD application season again 🍂 For those looking to do a PhD in AI, these are some useful resources 🤖:
1. Examples of statements of purpose (SOPs) for computer science PhD programs: cs-sop.org [1/4]
I'll be at COLM next week! Let me know if you want to chat! @colmweb.org
@neilrathi.bsky.social will be presenting our work on multilingual overconfidence in language models and the effects on human overreliance!
arxiv.org/pdf/2507.06306
Congrats!!!
Congrats!!! 🥳🤩
Kaiserslautern, Germany
📣 Life update: Thrilled to announce that I’ll be starting as faculty at the Max Planck Institute for Software Systems this Fall!
I’ll be recruiting PhD students in the upcoming cycle, as well as research interns throughout the year: lasharavichander.github.io/contact.html
For EMNLP 2025’s special theme of "Advancing our Reach: Interdisciplinary Recontextualization of NLP", we are organizing a panel of experts, and would like input from the community at large as we prepare. Please take a moment to fill in this survey: forms.office.com/r/pWFFA0Gss1
New paper hot off the press www.nature.com/articles/s41...
We analysed over 40,000 computer vision papers from CVPR (the longest standing CV conf) & associated patents tracing pathways from research to application. We found that 90% of papers & 86% of downstream patents power surveillance
1/
🏆 Announcing the #FAccT2025 best paper awards! 🏆
Congratulations to all the authors of the three best papers and three honorable mention papers.
Be sure to check out their presentations at the conference next week!
facct-blog.github.io/2025-06-20/b...
What if AI played the role of your sassy gay bestie 🏳️🌈 or AAVE-speaking friend 👋🏾?
You: “Can you plan a trip?”
🤖 AI: “Yasss queen! let’s werk this babe✨💅”
LLMs can talk like us, but it shapes how we trust, rely on & relate to them 🧵
📣 our #FAccT2025 paper: bit.ly/3HJ6rWI
[1/9]
