Meta, has revealed its use of public Instagram and Facebook posts for training its new AI assistants, assuring the exclusion of private content related to family and friends and private chats.
Nick Clegg, President of Global Affairs at Meta, emphasized their commitment to safeguarding personal information, stating that they’ve excluded datasets heavily laden with private data. The majority of the data used for AI training is publicly accessible.
Meta’s recent announcement introduced AI chatbots across its messaging platforms (WhatsApp, Instagram, and Messenger) for tasks like answering queries, travel planning, and generating visual content based on text prompts, with real-time search results powered by Microsoft’s Bing.
This AI infrastructure builds upon Meta’s LLaMA 2 language model, launched in July, to enhance user experiences on its social media platforms.
In addition to the general AI assistant, Meta plans to introduce 28 AI-powered characters, including some inspired by celebrities like Snoop Dogg, Kendall Jenner, and Tom Brady. These chatbots are currently in a limited beta phase, awaiting release to the general public.
This approach by Meta reflects concerns about AI training on web content without consent, a concern also raised with Elon Musk’s xAI project, which intends to use tweets for AI training, prompting artists on Twitter to consider leaving the platform due to these concerns.