Scholar
Terry Jingchen Zhang
Google Scholar ID: hFY4t8gAAAAJ
ETH Zurich
Citations & Impact
Citations
44
H-index
2
i10-index
2
Publications
10
Co-authors
16
list available
Publications
Resume (English only)
Academic Achievements
- Develops reasoning benchmarks grounded in research-level scientific publications; studies cross-domain transferability of emerging reasoning capabilities within (M)LLMs; proposes perturbation as a means to revive the utility of legacy benchmarks; introduces new signals for RL from internal/multi-agent feedback; explores key triggers for models to autonomously deceive/scheme/persuade other agents under more realistic or unrealistic scenarios.
Research Experience
- Attending ICML2026, Seoul, KR; Receives grants from Thinking Machines Lab and Open Philanthropy; Research directions span the science of evaluation, post-training, and alignment across multiple fronts.
Education
- PhD - Advisor not explicitly provided; MSc Interdisciplinary Science (CS and Physics), ETH Zurich, Switzerland (2024-2025), Thesis Advisors: Prof. Zhijing Jin, Prof. Bernhard Schölkopf; BSc Interdisciplinary Science (CS, Physics, and Chemistry), ETH Zurich, Switzerland (2023-2025), Thesis Advisor: Prof. Mrinmaya Sachan.
Background
- Research interests include improving AI agents' reasoning and alignment, especially in cooperative vs. competitive multi-agent scenarios that emulate real-world lab/corporate dynamics. Focuses on developing faithful evaluations to better inform post-training (reasoning-driven RL and alignment) using actionable interpretability (e.g., model diffing) and robustness probes (e.g., longitudinal analysis as a probe for contamination). In the AI4Science domain, interested in exoplanetary astrophysics, high-energy particle physics, and experimental chemistry (e.g., AI for spectroscopy and total synthesis).
Miscellany
- Personal tool recommendations: uv over conda or any other package manager; Asta is recommended as the best deep research tool; provides useful links regarding LM evaluation, post-training, etc.
