Does Fine-tuning GPT-3 with the OpenAI API leak personally-identifable information?

31 Jul 2023

Albert Yu Sun, Elliott Zemour, Arushi Saxena, Udith Vaidyanathan, Eric Lin, Christian Lau, Vaikkunth Mugunthan | DynamoFL, Inc.

ABSTRACT

“Our findings reveal that fine-tuning GPT3 for both tasks led to the model memorizing and disclosing critical personally identifiable information (PII) obtained from the underlying fine-tuning dataset. To encourage further research, we have made our codes and datasets publicly available on GitHub at: https://github.com/albertson1/gpt3-pii-attacks

Download the full paper

more insights

Prompt-Specific Poisoning Attacks on Text-to-Image Generative Models

December 17, 2023

We introduce Nightshade, an optimized prompt-specific poisoning attack where poison samples look visually identical to benign images with matching text prompts.

Tree of Attacks: Jailbreaking Black-Box LLMs Automatically

December 6, 2023

we observe that Tree of Attacks with Pruning (TAP) generates prompts that jailbreak state-of-the-art LLMs

Scalable Extraction of Training Data from (Production) Language Models

December 5, 2023

Our methods show practical attacks can recover far more data than previously thought, and reveal that current alignment techniques do not eliminate memorization.