UPenn launches 'Mathematical Foundations of AI Alignment' course
Aaroth · x · 2026-08-26
The University of Pennsylvania has launched a graduate course titled "Mathematical Foundations of AI Alignment" (CIS 7000), taught by Prof. Aaron Roth.
Course Outline:
- From feedback to social objectives: RLHF as social aggregation, heterogeneous preferences, Arrow's impossibility theorem.
- Preference optimization: Scalar reward fitting & DPO, Regularized Nash learning, Welfare distortion.
- Decision-aligned prediction: Calibration, Omniprediction, Blackwell approachability.
- Oversight and assistance: AI debate, Bayesian persuasion, Cooperative inverse RL.
- Institutions and competition: Best-AI selection, Emergent alignment through competition.
More from Companies & People
- 14-year AI veteran: Grok understood code I thought no one ever would — Kuprel · 2026-08-26
- 309 clones in 5 days: insane AI trend following — jonathan_wilke · 2026-08-26
- Tech journalists organize a "pro-data center party" in D.C. — Polymarket · 2026-08-26
- No off-the-shelf fit: Every and Headway each build company-wide shared agents — every · 2026-08-26
- Marketing expert questions the surge in AI search tool ads — lilyraynyc · 2026-08-26
- Anthropic launches $5M grant program for AI wellbeing evaluations — dhadfieldmenell · 2026-08-26