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

Maria Antoniak @mariaa.bsky.social · 3h

Umm… well… history shows… 😭

1 5

Reposted by Maria Antoniak

Bluesky @bsky.app · 1d

Quick Tip: Don’t want people getting pinged when you post or reply?

You can turn that off by heading to: Settings → Privacy and Security → Allow others to be notified of your posts and toggling it off.

A screenshot of “Allow others to be notified of your posts” with “No one” selected

190 530 3.2K

Reposted by Maria Antoniak

Antoine Bosselut @abosselut.bsky.social · 7h

If you're interested in doing a postdoc at @icepfl.bsky.social , there's still time to apply for the @epfl-ai-center.bsky.social postdoctoral fellowships.

Apart from this, I'm also recruiting postdocs in developing novel training algorithms for reasoning models and agentic AI.

1 2 5

Maria Antoniak @mariaa.bsky.social · 7h

"Finding Flawed Fictions: Evaluating Complex Reasoning in Language Models via Plot Hole Detection"
Kabir Ahuja et al. arxiv.org/abs/2504.11900

Figure 1: Example of FLAWEDFICTIONSMAKER (without the filtering step) in action that can be used to introduce plot holes in a plot hole-free story.

Maria Antoniak @mariaa.bsky.social · 7h

"Supposedly Equivalent Facts That Aren't? Entity Frequency in Pre-training Induces Asymmetry in LLMs" by Yuan He et al. arxiv.org/abs/2503.22362

Title, authors, and abstract of the paper

Figure 1: LLMs can exhibit asymmetry when recognising equivalent facts, often identifying facts from high-frequency to low-frequency entities but struggling with the inverse. Shown here is a working example from our tests with the OLMo2-13B model.

1 5

Maria Antoniak @mariaa.bsky.social · 7h

Inspired to share some papers that I found at #COLM2025!

"Register Always Matters: Analysis of LLM Pretraining Data Through the Lens of Language Variation" by Amanda Myntti et al. arxiv.org/abs/2504.01542

Figure 3: Change of accuracy from first to final checkpoint on individual benchmarks shown as a range, with grey indicating the first checkpoint and colours indicating the last checkpoint. The random-guess threshold is shown as a grey vertical line in cases where at least one model falls below it. Bars and legend shown in order of average accuracy.

1 8 20

Maria Antoniak @mariaa.bsky.social · 7h

This takes me back down an old rabbit hole...

"Translator Jay Rubin cut about 61 of 1,379 pages, including three chapters... These chapters contain plot elements not found elsewhere in the book.... some chapters were moved ahead of others, taking them out of the context of the original order... "

The Wind-Up Bird Chronicle - Wikipedia

en.wikipedia.org

Reposted by Maria Antoniak

Divya Shanmugam @dmshanmugam.bsky.social · 9h

I am on the job market this year! My research advances methods for reliable machine learning from real-world data, with a focus on healthcare. Happy to chat if this is of interest to you or your department/team.

1 11 21

Reposted by Maria Antoniak

arxiv cs.CL @arxiv-cs-cl.bsky.social · 13h

Federica Bologna, Tiffany Pan, Matthew Wilkens, Yue Guo, Lucy Lu Wang
LONGQAEVAL: Designing Reliable Evaluations of Long-Form Clinical QA under Resource Constraints
https://arxiv.org/abs/2510.10415

1 5

Reposted by Maria Antoniak

Maria Ryskina @mryskina.bsky.social · 1d

⭐ A thread for some cool recent work I learned about at #COLM2025, either from the paper presentations or from the keynotes!

1 6 13

Reposted by Maria Antoniak

Meredith Martin @mmvty.bsky.social · 1d

📣 New preprint! We know humans are biased against AI-creativity. But what about LLMs, now often judging creativity in various contexts? Do they replicate, transform, or amplify this bias? We tested it. Turns out: AI is 2.5X more biased against its own work than humans. arxiv.org/pdf/2510.08831 🧵

arxiv.org

3 24 60

Reposted by Maria Antoniak

chenhaotan.bsky.social @chenhaotan.bsky.social · 2d

What is the best way to find faculty job advertisements for CS/info students on the job market? It seems that students in my group are spending an unreasonable amount on this. I thought CRA is good, but seems not active any more.

3 4 12

Maria Antoniak @mariaa.bsky.social · 1d

I also had a great time at #COLM2025! I especially liked the long poster sessions (no need to rush through, plenty of time to see everything and chat with everyone) and single track talks.

Juan Diego Rodriguez @juand-r.bsky.social · 2d

#COLM2025 was one of my favorite conferences -- a really high fraction of interesting papers and people, but small enough to see everything!
Thank you to the organizers for putting it together!

2 19

Reposted by Maria Antoniak

Dustin Wright @dustinbwright.com · 1d

Which, whose, and how much knowledge do LLMs represent?

I'm excited to share our preprint answering these questions:

"Epistemic Diversity and Knowledge Collapse in Large Language Models"

📄Paper: arxiv.org/pdf/2510.04226
💻Code: github.com/dwright37/ll...

1/10

2 20 79

Maria Antoniak @mariaa.bsky.social · 1d

Yeah that’s the same route I was looking at and in total, 2+ hours even in the best case (middle of the day when there are somewhat frequent buses).

Maria Antoniak @mariaa.bsky.social · 1d

It’s so far! 😭

Maria Antoniak @mariaa.bsky.social · 1d

I believe that would take 2+ hours unless I’m missing something!

Maria Antoniak @mariaa.bsky.social · 2d

That’s good to know!

Maria Antoniak @mariaa.bsky.social · 2d

Yeah my current frustration is definitely also fueled by arriving from Montreal 😭

Maria Antoniak @mariaa.bsky.social · 2d

I’ve never owned a car, only occasionally renting or using a car share, for environmental and public health convictions. But Boulder might finally be the city to break me. Bad infrastructure mostly built for weekend warriors. Currently in a $90 Lyft from the airport because the hourly bus was full.

5 37

Maria Antoniak @mariaa.bsky.social · 2d

I tried the same query with @kagi.com and the results seem better to me! First hit is a set of opinionated reddit threads from r/houseplants.

Reposted by Maria Antoniak

Mel Andrews @bayesianboy.bsky.social · 6d

What problem is explainability/interpretability research trying to solve in ML, and do you have a favorite paper articulating what that problem is?

15 10 57

Maria Antoniak @mariaa.bsky.social · 2d

"come on pleeeease we only want to persuade you of good things! please please just let us persuade you a little bit! we promise only for GOOD things! come on!"

Maria Antoniak @mariaa.bsky.social · 2d

"nlp for social good"

Marjorie James Keenan @unenthusiast.com · 2d

In honour of spooky month, share a 4 word horror story that only someone in your profession would understand.

rm -rf ~/

hammancheez @hammancheez.bsky.social · 2d

"The chancellor approved it"

1 1 32

Reposted by Maria Antoniak

David Bau @davidbau.bsky.social · 3d

What does an LLM do when it translates from Italian "amore" to Spanish "amor" or French "amour"?

That's easy! (you might think) Because surely it knows: amore, amor, amour are all based on the same Latin word. It can just drop the "e", or add a "u".

2 2 34