Google researchers’ attack prompts ChatGPT to reveal its training data

A team of researchers primarily from Google’s DeepMind systematically convinced ChatGPT to reveal snippets of the data it was trained on using a new type of attack prompt which asked a production model of the chatbot to repeat specific words forever. 

Using this tactic, the researchers showed that there are large amounts of privately identifiable information (PII) in OpenAI’s large language models. They also showed that, on a public version of ChatGPT, the chatbot spit out large passages of text scraped verbatim from other places on the internet.

So not only are these things cases of mass copyright infringement, they also violate countless privacy laws.

Cool.

5 Comments

  1. 2023-12-01 4:23 pm
    • 2023-12-01 10:42 pm
      • 2023-12-02 10:13 am
  2. 2023-12-02 3:07 pm
    • 2023-12-02 11:34 pm