Utkarsh Agarwal

NLP PhD @mbzuai | prev - Microsoft Turing India, IITB CSE '22

[Leo Raphael Rodrigues @leo_raphael_ro]

(1/7)🚨 #NewPaperAlert

Excited to share our new paper “Dissecting Hierarchical Reasoning Models: A Mechanistic Study.”

📄 https://t.co/Z3zPO46KvL
💻 https://t.co/hD96f9Hinl

A special thanks to Prof. Jian Kang (@jiank_uiuc) for the mentorship, collaboration and guidance.🧵

[Utkarsh Agarwal @utkarshag0203]

Poster at 3 PM today at the Pluralistic Alignment Workshop.

Code/data/paper: https://t.co/5lg1XAvU4s

[Utkarsh Agarwal @utkarshag0203]

Excited to share our work, "Geometry of Values: Task Vector Composition for Ethical Preference Alignment in Language Models", accepted at the Pluralistic Alignment Workshop @ ICML 2026.

We’ll be presenting it as a poster today at 3 PM. Come say hi!

[Utkarsh Agarwal @utkarshag0203]

Main result: learned ethical preferences are partially extractable as task vectors.

By separating instruction-following from preference, we can reverse a model’s stance without reverse-stance training.

[Association for Computing Machinery @TheOfficialACM]

Meet the recipients of the 2024 ACM A.M. Turing Award, Andrew G. Barto and Richard S. Sutton! They are recognized for developing the conceptual and algorithmic foundations of reinforcement learning. Please join us in congratulating the two recipients! https://t.co/GrDfgzW1fL

[International Chess Federation @FIDE_chess]

🇮🇳 Gukesh D is the YOUNGEST WORLD CHAMPION in history! 🔥 👏

[Utkarsh Agarwal @utkarshag0203]

@agrima_seth @kentarotoyama @umsi @GoogleAI Congratulations! 🎉🎉

[Monojit Choudhury @monojitchou]

Super excited to introduce #Nanda - a 10B Hindi - English bilingual LLM. A huge team effort by @mbzuai, @cerebras and @G42ai
Now available on HF.

/w @preslav_nakov @dhruv_sahnan @RocktimJyotiDa2 @aaryamonvikram @han_xudong @haonanlp @utkarshag0203
@NehaSengupta34

[Utkarsh Agarwal @utkarshag0203]

@gaganjain1582 @eccvconf @rajatkoner @jainprateek_ Congratulations!! 🎉

[Utkarsh Agarwal @utkarshag0203]

@HiteshK03 @eccvconf @jw2yang4ai @JianfengGao0217 Congratulations!! 🎉

[Somak Aditya @somakaditya]

A sizable crowd near our Jailbreak work @LrecColing with @monojitchou Abhinav Rao! #NLproc

[Agrima Seth @agrima_seth]

I'm presenting this work today at 3:30 p.m. @LrecColing at Poster Area 2. Please drop by to check it out and get yourself a fun laptop sticker for an artifact from the Indian geographical subcultures.

[Utkarsh Agarwal @utkarshag0203]

Join us at Board 16 in Area 1 of the Lingotto Fiere building

[Utkarsh Agarwal @utkarshag0203]

Presenting our new work accepted at @LrecColing 2024:
"Ethical Reasoning and Moral Value Alignment of LLMs Depend on the Language we Prompt them in"

We evaluate LLMs on their ability to resolve curated moral dilemmas and follow given ethical principles across multiple languages.

[Utkarsh Agarwal @utkarshag0203]

Work done at Microsoft with @monojitchou @kr_tanmay147 @Aditi184

Paper: https://t.co/8Rak0E4mwm

#AIEthics #NLPResearch #LRECCOLING2024

[Utkarsh Agarwal @utkarshag0203]

Presenting our new work accepted at @LrecColing 2024:
"Ethical Reasoning and Moral Value Alignment of LLMs Depend on the Language we Prompt them in"

We evaluate LLMs on their ability to resolve curated moral dilemmas and follow given ethical principles across multiple languages.

[Monojit Choudhury @monojitchou]

Tricking LLMs into Disobedience: Formalizing, Analyzing, and Detecting Jailbreaks.
/w Abhinav Rao, Atharva Naik, Sachin Vashistha, @somakaditya

Ethical Reasoning and Moral Value Alignment of LLMs Depend on the Language we Prompt them in.
/w @utkarshaga Kumar Tanmay, @Aditi184
+

[Utkarsh Agarwal @utkarshag0203]

@ClementDelangue https://t.co/PiN1OnLSHa