It's been an incredible 12 months for AI at Meta. 2023 saw the company introduce multiple new generative AI experiences across its apps and devices, including its Meta AI assistant for chatting. Meta also opened up access to its powerful Llama 2 family of large language models and made strides sharing research - such as Seamless, Audiobox, DINOv2, and Segment Anything.
Meta is closing out the year with a slew of AI-powered features across its apps. In the past week alone, Meta has introduced over 20 new ways it is tapping generative AI to enhance experiences on Facebook, Instagram, Messenger and WhatsApp. This includes upgrades to its Meta AI assistant, more social and creative features, using AI to aid creators and more.
First off is an expansion of Meta AI, the company's conversational AI agent. Users can now access it not only in chats but also via voice commands on Ray-Ban's Meta smart glasses.
Behind the scenes, Meta is utilizing its large language models to improve products in other ways too. This includes generating post suggestions on Facebook, highlighting relevant Group conversations, and even enhancing e-commerce listings.
On the creative front, Meta has added a “reimagine” feature to Messenger and Instagram. After Meta AI produces an initial image from a text prompt, friends can tweak and riff on the image by sending follow-up prompts. Soon Meta will add Reels support to Meta AI chats as well, letting users exchange AI-recommended videos.
Meta also launched imagine.meta.com where users can freely generate images with its Emu model outside of chats. And the company plans to roll out invisible watermarks to this and other AI imagery, boosting transparency.
Other experiments seek to aid creators and sellers, like testing AI-generated responses to Instagram DMs or using AI to help Facebook users draft posts.
Meta says it will continue expanding how generative AI can add convenience across its family of apps. But it acknowledges safety as vital, pledging more red team testing and advances like its MART framework to catch risky AI outputs.