In an ambitious move to compete with OpenAI’s ChatGPT, Meta is introducing its own virtual assistant, Meta AI, capable of answering queries using Microsoft’s Bing search engine and creating images from textual commands. This launch marks a significant step in the increasingly competitive landscape of AI chatbots.
Powered by Meta’s expansive language model, Llama 2, Meta AI extends beyond simple text interaction. It can generate images using Emu, a newly developed image generator that Meta has trained on more than a billion pairs of photos and text, including content shared on Facebook or Instagram.
Initially available to a select group of US users across Facebook Messenger, Instagram, and WhatsApp, Meta AI can be integrated into group chats to assist with various tasks, such as planning travel. It will also be accessible via voice through Meta’s smart glasses, set to be released next month for US users.
The introduction of Meta AI places the social networking giant in direct competition with OpenAI’s ChatGPT-powered voice assistant and text-to-image generator. Despite trailing behind Microsoft, Alphabet, and Amazon in launching its own chatbot, Meta’s latest offering signals a strong response to ChatGPT.
Meta CEO Mark Zuckerberg announced Meta AI at an event today, which also saw the unveiling of the new Meta Quest 3 VR headset and a new model of smart glasses. However, the spotlight was on a series of generative AI updates, including two photo editing tools for Instagram.
Backdrop, one of these tools, can replace the background of a photo based on a text prompt. Restyle, the other tool, uses generative AI for creative effects, such as surrounding a person with puppies.
To avoid confusion between real and AI-generated imagery, Meta plans to watermark images created inside Meta AI, Backdrop, or Restyle. However, the specifics of this watermarking system have yet to be released.
Adding to its AI repertoire, Meta has also introduced a collection of chatbots modeled on approximately 30 celebrities, including tennis star Naomi Osaka and former football player Tom Brady. The celebrity chatbots, built on Llama 2, are available in beta on Facebook Messenger, Instagram, and WhatsApp.
Unlike competitors Google and OpenAI, who keep their latest AI models proprietary, Meta’s strategy emphasizes open-sourced machine-learning models. The company released the first Llama model to the public in February and followed up with the more advanced Llama 2 in July.
The integration of Llama into every Meta app and service exemplifies the benefits of open-sourcing AI models for large corporations, says Nathan Lambert, an AI researcher at Hugging Face. However, some critics, like Holly Elmore, are concerned about the potential risks associated with Meta’s open source AI strategy.
Meta AI isn’t the company’s first foray into AI assistants. In 2015, following the acquisition of several AI startups specializing in conversational AI, Meta introduced a virtual assistant named M. Despite its promising start, the majority of responses were generated by human workers rather than algorithms, leading to its quiet discontinuation in 2018.
With the launch of Meta AI, Mark Zuckerberg and his team are once again stepping into the world of AI assistants, aiming to make a more significant impact this time around.