Yong Zheng-Xin

Current: Computer Science Ph.D. @ Brown University
Past: Research Scientist Intern @ Meta AI, Research Collaborator @ Cohere Labs

prof_pic.jpg

I am a PhD student at Brown University advised by Prof. Stephen Bach. I am fortunate to have interned/collaborated at Meta (GenAI, FAIR) and Cohere Labs.

I currently work on reasoning models. My recent work:

  • We can monitor and even predict safety label before models finish thinking (preprint).
  • Scaling test-time compute helps cross-lingual reasoning (preprint).

I also work on safety alignment. I discovered that low-resource languages can jailbreak GPT-4 (⭑Best Paper Award, NeurIPS 2023 SoLaR Workshop; featured on New Scientist), which became the seminal work for multilingual red-teaming. My follow-up work used mechanistic interpretability to study crosslingual generalization of detoxification alignment training (EMNLP 2024 Findings) and adversarial attacks (NAACL 2025 Findings). We also recently released a survey on multilingual AI safety (preprint).

In addition, I work on multilingual LLMs and low-resource NLP. I co-developed Aya model (⭑Best Paper Award, ACL 2024) and worked on making speech models robust to accents (INTERSPEECH 2025). I also studied how pretrained LLMs can learn low-resource languages effectively through language adaptation (ACL 2023) and synthetic data (EMNLP 2024 Findings), as well as how they naturally mix languages (i.e. code-switching) (EMNLP 2023 CALCS, EMNLP 2023 Findings, ACL 2023 Findings).

Actively seeking full-time research roles in industry.


Featured work/preprints (see all)

  1. Yik Siu Chan* ,  Zheng-Xin Yong* ,  and  Stephen H. Bach
    arxiv preprint, 2025
  2. Zheng-Xin Yong ,  Beyza Ermis ,  Marzieh Fadaee , and 2 more authors
    arxiv preprint, 2025
  3. Zheng-Xin Yong ,  M. Farid Adilazuarda ,  Jonibek Mansurov , and 7 more authors
    arxiv preprint, 2025
  4. Samuele Poppi ,  Zheng-Xin Yong ,  Yifei He , and 4 more authors
    NAACL Findings, 2025
  5. Xiaochen Li* ,  Zheng-Xin Yong* ,  and  Stephen H Bach
    EMNLP Findings, 2024
  6. Ahmet Üstün* ,  Viraat Aryabumi* ,  Zheng-Xin Yong* , and 14 more authors
    ACL, 2024 (Best Paper Award)
  7. Zheng-Xin Yong ,  Cristina Menghini ,  and  Stephen Bach
    NeurIPS Workshop: Socially Responsible Language Modelling Research (SoLaR) , 2023 (Best Paper Award)

news

07 / 2025 Gave an invited talk at Google Multilinguality Reading Group.
05 / 2025 Gave an invited talk at MilaNLP lab.
05 / 2025 1 paper accepted! Work on understanding accent bias in ASR speech models was accepted to INTERSPEECH’25. Work was done during Meta internship.
02 / 2025 1 paper accepted! Work on cross-lingual finetuning attacks was accepted to NAACL’25 findings.
Work was done during Meta internship.
09 / 2024 4 papers accepted! LexC-Gen, SEACrowd, and crosslingual alignment were accepted to EMNLP. CVQA was accepted to NeurIPS.
08 / 2024 Aya Model paper received the ⭑Best Paper Award at ACL 2024.
07 / 2024 Gave an invited talk at London Data Week.
06 / 2024 Started research scientist internship at Meta AI (FAIR)!
05 / 2024 1 paper accepted! A Safe Harbor for AI Evaluation and Red Teaming is accepted to ICML.
02 / 2024 Released Aya model and dataset papers!
I also presented Aya multilingual safety research at Aya Grand Finale.
11 / 2023 Co-organized the tutorial on current status of NLP in South East Asia at AACL 2023.
10 / 2023 Low-Resource Languages Jailbreak GPT-4” received the ⭑Best Paper Award at NeurIPS 2023 Socially Responsible Language Modeling (SoLaR) workshop.
09 / 2023 Joined the Cohere For AI’s Responsible Deployment Team for Aya red-teaming.
08 / 2023 Served as the Area Chair (Multilingualism & Linguistic Diversity Track in EMNLP 2023).
05 / 2023 Media: Our code-switching paper was featured by Wired.
05 / 2023 3 papers accepted! BLOOM+1, BLOOMZ and code-switching survey were accepted to ACL 2023.
03 / 2022 2 papers accepted! T0 was accepted to ICLR (Spotlight). PromptSource was accepted to ACL Demo.
06 / 2021 Started PhD at Brown University.

Miscellaneous


✈️ I lived in 6 different countries (USA, UK, S.Korea, Argentina, India, and Germany) for at least four months each when I studied at Minerva University.

🕺 I love dancing. I used to teach a bit of Lindy Hop and salsa. Also learned the full K-pop choreo of “Let’s kill this love” for my undergrad graduation.

👃 Used to work on lung cancer diagnosis research (Google Science Fair 2016 Finalist; work published on Journal of Thoracic Disease 2016 and featured on IEEE Spectrum) before pivoting to AI.