Presented by
Cornell AI Alignment
A community of students and researchers conducting research and outreach to mitigate risks from advanced AI systems.
Hosted By
17 Went
📣 Guest Talk: Distortion of AI Alignment, by Prof. Paul Gölz
Registration
Past Event
About Event
Join Cornell AI Alignment for a guest talk by Prof. Paul Gölz on his NeurIPS 2025 paper, Distortion of AI Alignment: Does Preference Optimization Optimize for Preferences?
Modern AI systems are often aligned using methods like RLHF and DPO that learn from human preference data, but do these methods actually optimize for what people prefer? Prof. Gölz will discuss his work examining this question and what it means for how we think about preference optimization and AI alignment.
📅 Tuesday, September 8
🕠 5:30-6:30 PM
📍 CIS Building, Room 350
🍕 Food will be provided
Paper: https://arxiv.org/abs/2505.23749
Everyone is welcome!
Location
CIS Building 350
Presented by
Cornell AI Alignment
A community of students and researchers conducting research and outreach to mitigate risks from advanced AI systems.
Hosted By
17 Went