M29.3 CONNECT THE MECHANISM
Understand what training and deployment can reveal about data
Researchers got a chatbot to recite its training data by asking it to repeat one word forever. See how models leak, and how to add just enough noise to protect one person.
LESSON OVERVIEW15 min lesson
Lesson overview
Researchers got a chatbot to recite its training data by asking it to repeat one word forever. See how models leak, and how to add just enough noise to protect one person.
What you’ll explore
- Privacy risks arise from memorization, membership signals, extraction, and application data flows; differential privacy bounds a defined training-data influence while other controls address storage and access.
GO TO THE SOURCE
Original explanations, connected to the research.
Deep Learning with Differential Privacy (Abadi et al., 2016)Extracting Training Data from Large Language Models (Carlini et al., 2021)Extracting Training Data from Diffusion Models (Carlini et al., 2023)Scalable Extraction of Training Data from (Production) Language Models (Nasr et al., 2023)Membership Inference Attacks against Machine Learning Models (Shokri et al., 2017)The Algorithmic Foundations of Differential Privacy (Dwork & Roth, 2014)Simple Demographics Often Identify People Uniquely (Sweeney, 2000)Suggest a correction
A precise note can make an explanation better.
Choose the scene and describe what needs attention. Download a feedback file to share through a channel you already use. This page does not send feedback or connect you with a reviewer.