Hey, I'm Jasmine!

I'm trying to make AI development go well. Currently:

• I'm a MATS 9/9.1 Fellow under Alex Turner, working on training mitigations for evaluation gaming.
• Thinking about compute verification.

Previously, I built automated agent redteaming infra at Gray Swan AI, researched LLM honesty with the Center for AI Safety, and presented alignment benchmarks at NeurIPS and ICLR. I'm on leave from Cornell, where I studied CS & English as a Milstein and Telluride scholar. While there, I founded Cornell AI Alignment and Cornell's first AI safety class.

In high school, I really enjoyed tinkering with microfluidics, and published in Nature and analytical chemistry. Some other loves: persimmons, positive hyperstition, party hosting, writing, the 10% pledge 🔸, history of progress, outdoor physical challenges (half-marathon-ing, thru-hiking), tomato egg noodles, shoehorns, the Bay Area, my friends.

Please reach out and say hi (jasminexinzeli@gmail.com)!