Lightnews — Scholar-powered news

Maria Antoniak

@mariaa.bsky.social

10K followers 1.3K following 2K posts

☀️ Assistant Professor of Computer Science at CU Boulder 👩‍💻 NLP, cultural analytics, narratives, online communities 🌐 https://maria-antoniak.github.io 💬 books, bikes, games, art

maria-antoniak.github.io

Posts Media Videos Starter Packs

Pinned

Maria Antoniak @mariaa.bsky.social · Nov 20

some little bluesky tips 🦋

your blocks, likes, lists, and just about everything except chats are PUBLIC

you can pin custom feeds; i like quiet posters, best of follows, mutuals, mentions

if your chronological feed is overwhelming, you can make and pin make a personal list of "unmissable" people

15 53 240

Reposted by Maria Antoniak

David Mimno @dmimno.bsky.social · 3h

He had physical copies of at least six books at the podium and read passages from them. His main point was that regardless of whether you're more concerned about Big Risks or current harms, the "everything is great" position is untenable for anyone.

1 1 3

Reposted by Maria Antoniak

Thomas Davidson @thomasdavidson.bsky.social · 3h

There is one week left to apply to join us at Rutgers! We're hiring an Assistant Professor in Computational Sociology as part of a cluster of new hires in data science and AI.

Applications are due next Wednesday, 10/15.

Assistant Professor in Computational Sociology

The Department of Sociology at Rutgers University, New Brunswick, seeks applications for a tenure-track position at the Assistant Professor level specializing in Computational Sociology. The search i...

jobs.rutgers.edu

6 7

Maria Antoniak @mariaa.bsky.social · 3h

Another oral talk, this one by @stellali.bsky.social, discussing "PrefPalette: Personalized Preference Modeling with Latent Attributes."

Communities prefer different kinds of responses, which prioritize specific values. Aggregating preferences over communities would lose that signal.

#COLM2025

PrefPalette: Personalized Preference Modeling with Latent Attributes

Personalizing AI systems requires understanding not just what users prefer, but the reasons that underlie those preferences - yet current preference models typically treat human judgment as a black bo...

arxiv.org

1 8

Maria Antoniak @mariaa.bsky.social · 3h

At the end of the talk, he said he couldn't provide a full answer to that question but that we should each consider it. He said that he thinks there are potential upsides that he finds worth it, and that's why he's working on LMs. But he's not sure.

(paraphrasing a little bit from memory)

1 2

Maria Antoniak @mariaa.bsky.social · 3h

Now we're hearing a talk by @valentinhofmann.bsky.social about "Fluid Language Model Benchmarking."

Computerized adaptive testing is used for humans (like the GRE). This work adopts Item Response Theory from education to measure benchmark characteristics.

#COLM2025

Fluid Language Model Benchmarking

Language model (LM) benchmarking faces several challenges: comprehensive evaluations are costly, benchmarks often fail to measure the intended capabilities, and evaluation quality can degrade due to l...

arxiv.org

1 7

Maria Antoniak @mariaa.bsky.social · 3h

Question from the audience (didn't catch the name): Why are computer scientists the ones who should solve this problem? They lack expertise, and there are other people who have been studying these kinds of harms for a very long time.

#COLM2025

2 7

Maria Antoniak @mariaa.bsky.social · 3h

"What problems you're scared of depend on how good you think the LLMs will get"

"Please be willing to change your mind."

"This is COLM. We made the models, it's our job to fix it. How are you going to change your research agenda?"

#COLM2025

2 9

Maria Antoniak @mariaa.bsky.social · 3h

"I'm not arguing that everyone should work on all of these problems, I don't know how to work on this problem. But we should work on them scientifically and on a spectrum of problems."

Argues that these different kinds of risks are NOT just distractions from each other.

#COLM2025

2 6