AI has had an amazing year. We shared our Llama family of large language models with you, unveiled new AI experiences across our apps and devices, and released research discoveries like Emu Video and Emu Edit that will unlock new features in our products next year. With AI developments in content creation, speech, and multimodality that will allow us to provide fresh, imaginative, and immersive applications, we are eager for what lies ahead in store for us next year. We’re sharing upgrades to some of our most popular AI experiences today, along with new features you can find in all of our app family members.
The Development of Artificial Intelligence
Our virtual assistant, Meta AI, can provide photorealistic visuals, respond to queries, and more. We’re improving it to be more beneficial by providing more thorough mobile responses and precise search result summaries. We have even improved the likelihood that you will receive a good response to a larger variety of queries. To communicate with Meta AI, open a new message on one of our messaging platforms and choose “Create an AI chat,” or enter “@MetaAI” in a group chat and then specify the task you’d like the assistant to help you with. Additionally, you can use your Ray-Ban Meta smart glasses to say “Hey Meta.”
Additionally, Meta AI is now assisting you outside of conversations. It’s working hard in the background to improve the usability and enjoyment of our product experiences on Facebook and Instagram. With the help of Meta AI’s extensive language model technology, users in a number of English-speaking markets can choose from AI-generated post comment and community chat subject ideas in groups, as well as search results and improved product copy in stores. Additionally, it powers envision with Meta AI, a brand-new stand-alone experience for artistic enthusiasts.
With friends, create and riff on images
Imagine, our text-to-image generating option that allows you to generate and share images on the fly, is one of Meta AI’s most widely utilized services across our messaging apps. We’re constantly thinking of ways to make our AIs even more entertaining and sociable, which is why we’re thrilled to introduce reimagine, a new feature to Meta AI on Instagram and Messenger. In group chat, it operates as follows: With a simple text prompt, your buddy can riff on the image that Meta AI has generated and shared, and Meta AI will create a completely new image if they continue to press and hold on it. You may now chuckle while you kick photos back and forth.
Learn New Things Using Reels in Meta AI
Sometimes it’s not enough to just read about something; you also need to be able to see or experience it visually. You can accomplish precisely that with reels. They’re fantastic for finding inspiration, interacting with creators, and finding new material. Additionally, we are currently beginning to implement Reels in Meta AI discussions.
Let’s say you and your friends are organizing a trip to Tokyo. You may ask Meta AI to share Reels of the top attractions and suggest the best places to visit, helping you determine which attractions are a must-see.
This is just the beginning of how we plan to develop deeper app-wide interactions to gradually make Meta AI a more sophisticated personal assistant.
Improving Facebook Experience
Additionally, we’re still innovating on Facebook to support users with common experiences by facilitating expression and discovery in a way that has never been possible before.
We’re investigating ways to leverage Meta AI underneath the hood to assist you in using AI to compose a catchy introduction for your Facebook Dating profile, modify your own Feed posts, or even build the ideal birthday greeting to share with your best friend.
Additionally, we are investigating methods for creating and sharing AI-generated photos on Facebook that are simple, including converting photographs from landscape to portrait orientation using Meta AI to make sharing them to Stories easier.
We’re also experimenting with leveraging Meta AI to recommend subjects for new chats and surface pertinent information in Groups so you don’t miss out on the important discussions. This will help you stay involved in your communities. We’re investigating the use of Meta AI on Marketplace to assist customers in finding similar or substitute products with ease and learning more about the things they’re thinking about purchasing. Additionally, we’re experimenting with Meta AI to enhance our friend, page, group, and marketplace search functions.
Supporting Artists in Replying to Their Fans
Our goal is to provide creators with generative AI tools so they can work more productively and engage with more members of their community. We’re beginning to test recommended replies in direct messages (DMs) to assist creators in engaging with their audiences more quickly and effortlessly, building on our previous changes on Instagram. Meta AI will begin working in the background to generate pertinent replies for review depending on the tone and substance of messages that producers in the test open in their direct messages.
Imagine Using Meta AI to Experience
We’ve loved hearing from users about how they use Meta AI’s text-to-image generating function, imagine, to create amusing and imaginative conversation material. Imagine will now be accessible outside of conversations as well; users in the US can begin using it at imagine.meta.com. using the help of technologies from Emu, our image foundation model, you may create photos using this stand-alone experience for creative hobbyists. Even though our messaging system is meant to foster more light hearted, back-and-forth exchanges, you can now make free online photographs as well.
Speaking With Our AIs
We’re not merely developing meta AI. Based on your input, we’re also enhancing our other AIs. More of our AIs will soon be able to search. Since their debut, Bru and Perry, two of our AIs that specialize in sports, have been providing responses that are fueled by Bing. We are currently extending this feature to Luiz, Coco, Lorena, Tamika, Izzy, and Jade as well.
Furthermore, we’re testing a new feature that adds long-term memory to some AIs so that the knowledge they gain from your discussion stays with them after it ends. This implies that you can resume where you left off when you visit a certain AI. Our objective is to introduce the possibility of more meaningful interactions and expanded conversational capacities to your conversations with Billie, Carter, Scarlett, Zach, Victor, Sally, and Leo, among other AIs. Additionally, you can always delete the chat history you’ve had with our AIs. For additional information, see our Generative AI Privacy Guide.
Open a new message on WhatsApp, Messenger, or Instagram and choose “Create an AI chat” to have a conversation with our AIs. Anyone in the US can now access them.
Red Teaming, Invisible Watermarking, and Upcoming Events
We are dedicated to creating products that are built responsibly and with safety in mind, and we recognize the value of openness when it comes to the material that artificial intelligence produces. Numerous photographs produced by our tools show that artificial intelligence (AI) was used to lower the likelihood that they would be mistaken for human-generated content. We’ll enhance transparency and traceability in the upcoming weeks by incorporating invisible watermarking into the Imagine with Meta AI experience.
To apply the invisible watermark, a deep learning model is used. Even though the invisible watermark is invisible to the human eye, it can be found using a matching model. It resists standard image modification techniques like cropping, adjusting colour (brightness, contrast, etc.), taking screen grabs, and more. In the future, we hope to add invisible watermarking to a large number of our products that use AI-generated photos.
Additionally, we’re still making investments in red teaming, which has long been ingrained in our culture. We pressure test our generative AI research and features that use large language models (LLMs) with prompts that we anticipate could produce dangerous outcomes as part of that effort.
We recently presented Multi-round Automatic Red-Teaming (MART), an approach to enhance LLM safety that uses automatic iterative adversarial red teaming to train an adversarial and target LLM. We are attempting to continuously red team and enhance safety by integrating the MART framework into our AIs.
Lastly, we keep hearing from people about their impressions of our AIs, particularly Meta AI.
We have only begun to explore the potential of artificial intelligence. Await further updates in the upcoming year.
