Erik Brinkman

• Created AI policy guidelines and governance for safe LLM model deployment

• Enabled release of llama3 through red teaming to measure and fine-tune against adversarial risks [github.com/meta-llama/llama3]
• Pioneered applying high dimensional Bayesian optimization towards finding system level adversarial prompts for SOTA models
• Designed and developed key metrics for hyper-spherical point diversity, which were used to understand relative coverage of prompts and unsafe responses for model evaluation
• Developed SOTA calibration-based recommendation based fairness metrics, and applied them to internal systems.

Understanding Financial Market Behavior through Empirical Game-Theoretic Analysis