Moritz Hardt

Principal Researcher@Max-Planck-Institute for Intelligent Systems, Max-Planck Institute·德国·OpenReview

研究方向

language models · algorithmic collective action · performative prediction · test-time training · benchmarking · strategic classification · algorithmic fairness · adaptive data analysis · differential privacy

Moritz Hardt

Train-before-Test Harmonizes Language Model Rankings

Good Allocations from Bad Estimates

Learning on the Job: Test-Time Curricula for Targeted Reinforcement Learning

Answer matching outperforms multiple choice for LLM evaluations

Scaling Open-Ended Reasoning to Predict the Future

Training on the Test Task Confounds Evaluation and Emergence

How Benchmark Prediction from Fewer Data Misses the Mark

Lawma: The Power of Specialization for Legal Annotation

Monoculture or Multiplicity: Which Is It?

Limits to scalable evaluation at the frontier: LLM as judge won’t beat twice the data

Policy Design in Long-run Welfare Dynamics