Conrad Feng

name

Conrad Feng / cfeng204 / conrad204

email

conrad.feng [at] emory [dot] edu

looking for

internships, research, hackathons

work

interpretability research, ai safety, bioinformatics

main skills

python · pytorch · mech interp · llm evals · mcp servers · frontend/design · bioinformatics

current

ai safety research: mech interp w/ professor Ali Emami

resume

Conrad_Resume_9_25_2026.pdf

publications

1. Probing Evaluation-Awareness in Frontier Language ModelsC. Feng, A. Rao, M. Liang — NeurIPS 2025
2. Sparse Features Are Not Enough: Compositional Structure in Dictionary LearningJ. Park, C. Feng — ICML 2025
3. A Calibration Account of the Logit LensC. Feng, S. Okafor — ICLR 2026
4. Circuit-Level Signatures of Evaluation GamingC. Feng, A. Rao — preprint 2026
5. When Models Know They Are Being TestedD. Mehta, C. Feng — TMLR 2024