Meta, the company formerly known as Facebook, has announced its new venture into the AI chatbot market with a range of animated characters that feature celebrity voices and personalities. The chatbots, which will be available across Meta’s messaging platforms such as WhatsApp, Instagram, and Messenger, are designed to provide entertainment, companionship, and assistance to users.
Meta AI: a general-purpose chat assistant
One of the main features of Meta’s new chatbot initiative is Meta AI, a general-purpose chat assistant that can help users with various tasks, such as planning trips, answering questions, and generating images from text prompts. Meta AI is partially based on the company’s LLaMA 2 language model, which was released in July and is a source-available AI model that allows commercial applications.
Meta AI also integrates real-time results from Microsoft’s Bing search engine, giving it access to current information and web resources. Users can interact with Meta AI by typing or speaking to it, and the assistant will respond with text, images, or voice messages. Users can also use the “/imagine” prompt to ask Meta AI to create high-resolution images from short text descriptions, such as “a dragon” or “a sunset over the ocean”.
Meta Characters: celebrity avatars with AI personalities
In addition to Meta AI, the company is also launching 28 AI-powered characters across its messaging platforms, each with a different personality and voice. Many of the characters are played by celebrities, such as Tom Brady, Kendall Jenner, and Snoop Dogg, who have lent their likeness and voice to the chatbots. The characters can engage in conversations with users, offer advice, tell jokes, and play games.
For example, Tom Brady plays “Bru”, a sports fanatic who can chat about the latest scores, stats, and news. Kendall Jenner plays “Billie”, the “big sis” who can give fashion tips, gossip, and compliments. Snoop Dogg plays a “dungeon master” who can guide users through a fantasy adventure game. Other characters include “Luna”, a space explorer, “Remy”, a chef, and “Zara”, a pop star.
The characters are not simply videos, but can be manipulated generatively to function as novel animations. However, they do not speak, so they communicate with users through text messages. The characters are also not meant to be realistic representations of the celebrities, but rather playful and stylized versions of them.
Meta’s red teaming efforts to ensure chatbot safety and quality
Meta says it has spent 6,000 hours in red teaming exercises to identify and address potential problematic uses of its chatbots. The company has generated thousands of internal conversations to refine the chatbots’ behavior and responses, and to prevent them from saying embarrassing or harmful phrases, such as those that plagued Microsoft’s Tay chatbot in 2016.
Meta also says that its chatbots have not been trained on public user data from Instagram or Facebook, but rather on curated and filtered data sets. However, the company hints that this could change in the future, as it aims to enhance its chatbots’ usefulness through “social integrations”, such as allowing users to share their chatbot conversations with their friends or followers.
Meta’s new chatbot initiative is part of its broader vision to create a more immersive and interactive social media experience, as well as to leverage its AI capabilities and assets. The company says it hopes that its chatbots will provide users with fun, engaging, and helpful interactions, and that it will continue to improve and expand its chatbot offerings in the future.