Meta Platforms is training parts of its new artificial intelligence (AI) assistant with public Facebook (FB) and Instagram (IG) posts while safeguarding users’ data privacy.
At the Connect conference, Meta CEO Mark Zuckerberg shifted focus from augmented and virtual reality (AR/VR) to AI.
During the event, Meta Global Affairs President Nick Clegg revealed no usage of private chats on their messaging services. Clegg used LinkedIn as an example of a website whose content the social media giant did not use for privacy reasons.
Major tech companies, including Meta, OpenAI, and Google, have received criticism for using information online without approval for training their AI.
While facing copyright lawsuits, the companies are exploring ways to manage private or copyrighted content that their AI may reproduce.
Zuckerberg disclosed that Meta AI was developed from a custom model based on the firm’s large language model (LLM) Llama 2. Meta has also introduced its new AI image generator and Emu.
Partnering with Microsoft’s Bing, Meta AI will provide text, audio, imagery, and access to the latest information.
Clegg said the public posts on FB and IG used to train the AI assistant included text and images. He added that the tech giant employed safety measures on Meta AI’s content generation—for instance, a ban on producing photo-realistic images of well-known individuals.
According to a Meta spokesperson, the company has added new terms of service prohibiting users from creating content that violates privacy and intellectual property rights to prevent copyright issues.
Meta AI Chatbot to Launch Across Platforms
Meta officially participates in the AI chatbot wars as it plans to launch its AI assistant and characters on WhatsApp, Instagram, and Messenger.
The company is releasing Meta AI and a set of AI characters based on celebrities like MrBeast and Charli D’Amelio. It is rolling out its first set of 28 AI characters, which primarily includes high-profile individuals, across its messaging platforms.
The tech giant’s AI will have similar features to chatbots such as OpenAI’s ChatGPT and Anthropic’s Claude. Meta expects its AI assistant to help users with everything, from planning trips in a group chat to obtaining information for questions commonly asked on search engines.
The AI can also produce images like Midjourney or OpenAI’s DALL-E through the prompt “/image.” The image generator is free to use, as with all the tech giant’s AI features disclosed this week.
