In recent updates, tech giant Meta revealed that its latest AI assistant is trained using public posts from Facebook and Instagram. The company insists that user confidential data remains secure throughout the process.
Private posts and personal conversations from its platforms have been marked off-limits in preparing the AI training dataset. Nick Clegg, the Global Business President of Meta, emphasized that they strictly avoid datasets containing any private information.
According to Mr. Clegg, publicly available data forms the majority of sources used for training AI helpers. Meta’s recent announcement involved powerful AI chatbots for platforms like WhatsApp, Instagram, and Messenger. These chatbots are designed to handle a wide range of user tasks, from responding to queries to generating images from text prompts. Furthermore, they will deliver live search results powered by Microsoft’s Bing search engine.
The chatbots aren’t available to the general public yet, as they are still under beta testing. Meta officials reinforced that their approach of only using public posts to train AI falls in line with standard industry practices.
Enhancing the social experience, Meta is also bringing 28 AI-powered figures to their messaging platforms. Some of these characters are styled after celebrities like Snoop Dogg, Kendall Jenner, and Tom Brady.
To reinforce its stance on privacy, Meta went on record stating private chats from its messaging services do not serve as training data for AI models. Extra steps have been taken to sieve out any sort of private details from the public datasets used in training.
Such public revelations come at a time when tech companies have faced backlash for using web-scraped data without proper permits to train their AI models. These models require huge data volumes to summarize info and create related imagery.
Thus, the companies are now in a tight spot regarding differentiating between public, private, and copyrighted materials – all while facing lawsuits for supposed copyright infringement.
During the Meta’s annual Connect conference this week, these AI advancements were given more limelight than augmented or virtual reality developments, the usual focal point of these conferences.
AI assistants such as the one Meta introduced are created using a refined model based on the powerful Llama 2 large language model. Additionally, a new model called Emu that generates images in response to text prompts was also unveiled.
