Distill to Detect: Exposing Stealth Biases in LLMs through Cartridge Distillation
ICML 2026 Workshops · TAIGR, Mech Interp, CoLorAI, AI4Good
Abhinav Chinta
I'm a master's student in Computer Science at Stanford University, advised by Amin Saberi. I also work closely with Azalia Mirhoseini and Monica Lam. My research focuses on reasoning, post-training, and interpretability in large language models.
I'm currently a research scientist intern at Foundation AI (Cisco). Before Stanford, I received my B.S. in Computer Science from the University of Illinois Urbana-Champaign, where I worked with Dilek Hakkani-Tür and Heng Ji.
Contact: achinta [at] stanford.edu
News
- Jun 2026
- Joined Foundation AI (Cisco) as a research scientist intern.
- May 2026
- Distill to Detect accepted to four ICML 2026 workshops.
- Sep 2025
- Started my M.S. in Computer Science at Stanford University.
- May 2025
- Premise-Augmented Reasoning Chains accepted to ICML 2025.
Publications
Premise-Augmented Reasoning Chains Improve Error Identification in Math Reasoning with LLMs
ICML 2025
Unsupervised Human Preference Learning
EMNLP 2024
Democratizing LLMs: An Exploration of Cost-Performance Trade-offs in Self-Refined Open-Source Models
EMNLP 2023 · Findings
* denotes equal contribution
Experience
- Foundation AI (Cisco) · Research Scientist Intern Jun 2026 – Present
- Sandia National Laboratories · Research Intern May 2024 – Aug 2025
- ConvAI Lab (UIUC) · Research Assistant Aug 2023 – Aug 2025
- BLENDER Lab (UIUC) · Research Assistant Jun 2023 – Aug 2025
- Nference · Machine Learning Intern May 2023 – Aug 2023
- Jane Street · SEE Trading Fellow May 2023
Education
- Stanford University · M.S. in Computer Science 2025 – 2027
- University of Illinois Urbana-Champaign · B.S. in Computer Science 2021 – 2025