Lightnews — Scholar-powered news

Reposted by Ana Marasović

Daniel Brown @daniel-brown.bsky.social · 9h

Can you trust your reward model alignment scores?
New work presented today at the COLM Workshop on Socially Responsible Language Modelling Research led by Purbid Bambroo and in collaboration with @anamarasovic.bsky.social that probes LLM preference test sets for redundancy and inflated scores.

1/8

1 1 1

Reposted by Ana Marasović

Ana Marasović @anamarasovic.bsky.social · 1d

📣Tomorrow at #COLM2025:

1️⃣ Purbid's 𝐩𝐨𝐬𝐭𝐞𝐫 at 𝐒𝐨𝐋𝐚𝐑 (𝟏𝟏:𝟏𝟓𝐚𝐦-𝟏:𝟎𝟎𝐩𝐦) on catching redundant preference pairs & how pruning them hurts accuracy; www.anamarasovic.com/publications...

2️⃣ My 𝐭𝐚𝐥𝐤 at 𝐗𝐋𝐋𝐌-𝐑𝐞𝐚𝐬𝐨𝐧-𝐏𝐥𝐚𝐧 (𝟏𝟐𝐩𝐦) on measuring CoT faithfulness by looking at internals, not just behaviorally

1/3

1 3 12

Ana Marasović @anamarasovic.bsky.social · 1d

As in a version of this talk a few months back, I'll cover our #EMNLP2025 unlearning-based approach for measuring CoT parametric faithfulness led by @mtutek.bsky.social. Also presented by @boknilev.bsky.social in the INTERPLAY workshop (10:20-12pm)!

arxiv.org/abs/2502.14829

3/3

Measuring Chain of Thought Faithfulness by Unlearning Reasoning Steps

When prompted to think step-by-step, language models (LMs) produce a chain of thought (CoT), a sequence of reasoning steps that the model supposedly used to produce its prediction. Despite much work o...

arxiv.org

3

Ana Marasović @anamarasovic.bsky.social · 1d

A new addition to my talk is @fatemehc.bsky.social's ongoing work on 𝑎𝑢𝑡𝑜𝑚𝑎𝑡𝑒𝑑 𝑑𝑖𝑠𝑐𝑜𝑣𝑒𝑟𝑦 𝑜𝑓 𝑠𝑝𝑎𝑟𝑠𝑒 𝑓𝑒𝑎𝑡𝑢𝑟𝑒 𝑐𝑖𝑟𝑐𝑢𝑖𝑡𝑠 with 𝐠𝐫𝐚𝐩𝐡 𝐬𝐤𝐞𝐥𝐞𝐭𝐨𝐧𝐢𝐳𝐚𝐭𝐢𝐨𝐧

2/3

1 3

Ana Marasović @anamarasovic.bsky.social · 1d

📣Tomorrow at #COLM2025:

1️⃣ Purbid's 𝐩𝐨𝐬𝐭𝐞𝐫 at 𝐒𝐨𝐋𝐚𝐑 (𝟏𝟏:𝟏𝟓𝐚𝐦-𝟏:𝟎𝟎𝐩𝐦) on catching redundant preference pairs & how pruning them hurts accuracy; www.anamarasovic.com/publications...

2️⃣ My 𝐭𝐚𝐥𝐤 at 𝐗𝐋𝐋𝐌-𝐑𝐞𝐚𝐬𝐨𝐧-𝐏𝐥𝐚𝐧 (𝟏𝟐𝐩𝐦) on measuring CoT faithfulness by looking at internals, not just behaviorally

1/3

1 3 12

Ana Marasović @anamarasovic.bsky.social · 2d

Sad: Can't go to CoLM because of immigration.

Happy: Well, at least I can mountain bike during the fall break in prime SLC MTB weather.

Sad: Comes down with a cold.

☹️☹️☹️☹️☹️☹️

7

Ana Marasović @anamarasovic.bsky.social · 2d

I had a great time chatting with Jekaterina and Malikeh. This episode is like a tour of all the things I've been studying lately!

Women in AI Research - WiAIR @wiair.bsky.social · 2d

🎙️ New Women in AI Research episode out now!
This time, we sit down with @anamarasovic.bsky.social to unpack some of the toughest questions in AI explainability and trust.

🔗 Watch here → youtu.be/xYb6uokKKOo

youtu.be

1

Reposted by Ana Marasović

Women in AI Research - WiAIR @wiair.bsky.social · 2d

🎙️ New Women in AI Research episode out now!
This time, we sit down with @anamarasovic.bsky.social to unpack some of the toughest questions in AI explainability and trust.

🔗 Watch here → youtu.be/xYb6uokKKOo

youtu.be

1 1 1

Ana Marasović @anamarasovic.bsky.social · 2d

Happening today! #COLM2025

Ana Marasović @anamarasovic.bsky.social · 4d

Honored 🎷🎸🥁 𝗠𝗶𝘅𝗔𝘀𝘀𝗶𝘀𝘁 🎷🎸🥁 was selected as the #COLM2025 oral spotlight. Go check out @mclemcrew.bsky.social's 𝐭𝐚𝐥𝐤 on 𝐖𝐞𝐝 (𝐎𝐜𝐭 𝟖) at 𝟑:𝟑𝟎𝐩𝐦 in 517BC and 𝐩𝐨𝐬𝐭𝐞𝐫 from 𝟒:𝟑𝟎-𝟓:𝟑𝟎 in 710!

MClem @mclemcrew.bsky.social · 4d

Mega stoked to attend #COLM25 this week and present our work, MixAssist, on Wednesday!

@anamarasovic.bsky.social sadly can't make it 😭, but hit me up if you'd like to chat about audio language models, music mixing, or anything else regarding music and audio!

Reposted by Ana Marasović

Yonatan Belinkov ✈️ COLM2025 @boknilev.bsky.social · 3d

In #Interplay25 workshop, Friday ~11:30, I'll present on measuring *parametric* CoT faithfulness on behalf of
@mtutek.bsky.social , who couldn't travel:
bsky.app/profile/mtut...

Later that day we'll have a poster on predicting success of model editing by Yanay Soker, who also couldn't travel

1 1 4

Reposted by Ana Marasović

Martin Tutek @mtutek.bsky.social · 3d

I won't be at COLM, so come see Yonatan talk about our work on estimating CoT faithfulness using machine unlearning!

Check out the thread for the (many) other interesting works from his group 🎉

Yonatan Belinkov ✈️ COLM2025 @boknilev.bsky.social · 3d

In #Interplay25 workshop, Friday ~11:30, I'll present on measuring *parametric* CoT faithfulness on behalf of
@mtutek.bsky.social , who couldn't travel:
bsky.app/profile/mtut...

Later that day we'll have a poster on predicting success of model editing by Yanay Soker, who also couldn't travel

1 3

Ana Marasović @anamarasovic.bsky.social · 4d

arxiv.org/abs/2507.06329

MixAssist: An Audio-Language Dataset for Co-Creative AI Assistance in Music Mixing

While AI presents significant potential for enhancing music mixing and mastering workflows, current research predominantly emphasizes end-to-end automation or generation, often overlooking the collabo...

arxiv.org

Ana Marasović @anamarasovic.bsky.social · 4d

Honored 🎷🎸🥁 𝗠𝗶𝘅𝗔𝘀𝘀𝗶𝘀𝘁 🎷🎸🥁 was selected as the #COLM2025 oral spotlight. Go check out @mclemcrew.bsky.social's 𝐭𝐚𝐥𝐤 on 𝐖𝐞𝐝 (𝐎𝐜𝐭 𝟖) at 𝟑:𝟑𝟎𝐩𝐦 in 517BC and 𝐩𝐨𝐬𝐭𝐞𝐫 from 𝟒:𝟑𝟎-𝟓:𝟑𝟎 in 710!

MClem @mclemcrew.bsky.social · 4d

Mega stoked to attend #COLM25 this week and present our work, MixAssist, on Wednesday!

@anamarasovic.bsky.social sadly can't make it 😭, but hit me up if you'd like to chat about audio language models, music mixing, or anything else regarding music and audio!

1 1 12

Reposted by Ana Marasović

MClem @mclemcrew.bsky.social · 4d

Mega stoked to attend #COLM25 this week and present our work, MixAssist, on Wednesday!

@anamarasovic.bsky.social sadly can't make it 😭, but hit me up if you'd like to chat about audio language models, music mixing, or anything else regarding music and audio!

1 3

Reposted by Ana Marasović

Women in AI Research - WiAIR @wiair.bsky.social · 7d

Our new guest at #WiAIRpodcast is @anamarasovic.bsky.social
(Asst prof @ University of Utah , Ex @ Allen AI). We'll talk with her about faithfulness, trust and robustness in AI.
The episode is coming soon, don't miss:
www.youtube.com/@WomeninAIRe...

#WiAIR #NLProc

1 2

Reposted by Ana Marasović

Lucy Lu Wang @lucylw.bsky.social · Sep 10

Pls help spread the word: We are hiring! 🌲☕✨
@ischool.uw.edu is recruiting two tenure-track assistant profs in AI. Areas include human-AI collaboration, AI & society, comp social science, NLP, and AI applications.

Apply by Nov 15, 2025: apply.interfolio.com/171020

Happy to answer questions!

Apply - Interfolio {{$ctrl.$state.data.pageTitle}} - Apply - Interfolio

apply.interfolio.com

1 13 22

Ana Marasović @anamarasovic.bsky.social · Sep 6

Exciting, congrats!! Can't wait to hear about what you built and how artists are using it 🤖🎵

2

Reposted by Ana Marasović

MClem @mclemcrew.bsky.social · Sep 6

Beyond excited to share that I’ll be joining Universal Audio as an LLM Applications Engineer!

So stoked to continue working on co-creative AI elements of music production in this role 🎹

S/O to @anamarasovic.bsky.social for teaching me all things NLP these past few years!

1 1 4

Ana Marasović @anamarasovic.bsky.social · Aug 29

Learned a ton working on this survey. Hope it's helpful for you too, check it out! Feedback welcome 🙏

kennethmarino.com/computeruse/...

Kenneth Marino @kennethmarino.bsky.social · Aug 29

Super excited that the Computer Use survey I've been working on w/ @anamarasovic.bsky.social for a while now is ready! Originally we were planning on a more traditional survey paper but as more surveys came out we decided on an interactive website survey.

2 3 6

Reposted by Ana Marasović

Kenneth Marino @kennethmarino.bsky.social · Aug 29

Super excited that the Computer Use survey I've been working on w/ @anamarasovic.bsky.social for a while now is ready! Originally we were planning on a more traditional survey paper but as more surveys came out we decided on an interactive website survey.

1 1 1

Ana Marasović @anamarasovic.bsky.social · Aug 21

Check @mtutek.bsky.social's thread!

Martin Tutek @mtutek.bsky.social · Aug 21

Thrilled that FUR was accepted to @emnlpmeeting.bsky.social Main🎉

In case you can’t wait so long to hear about it in person, it will also be presented as an oral at @interplay-workshop.bsky.social @colmweb.org 🥳

FUR is a parametric test assessing whether CoTs faithfully verbalize latent reasoning.

1 4

Reposted by Ana Marasović

Martin Tutek @mtutek.bsky.social · Aug 21

Thrilled that FUR was accepted to @emnlpmeeting.bsky.social Main🎉

In case you can’t wait so long to hear about it in person, it will also be presented as an oral at @interplay-workshop.bsky.social @colmweb.org 🥳

FUR is a parametric test assessing whether CoTs faithfully verbalize latent reasoning.

1 3 13

Ana Marasović @anamarasovic.bsky.social · Aug 20

Congrats!!

Ana Marasović @anamarasovic.bsky.social · Aug 20

Congrats!!

1

Ana Marasović @anamarasovic.bsky.social · Aug 15

Congrats!!

1 1