Google researchers' attack prompts ChatGPT to reveal its training data – OSnews
Google DeepMind demonstrated that ChatGPT retains significant personally identifiable information and verbatim copyrighted text from its training data. This revelation highlights critical privacy and copyright risks inherent in current large language models. For open data, it underscores the urgent need for stricter data governance and transparency in AI training pipelines.
Source: osnews.comPublished on 2023-12-02