Meta Company announced the launch of MetaAI and released a series of personalized artificial intelligence characters in its products. Given the sheer scale of the company's social network, its AI assistant may give most people their first taste of generative artificial intelligence.
Meta CEO Mark Zuckerberg emphasized that Meta's strategy involves creating different AI products for different uses, rather than a single flagship chatbot. Meta’s conversational bots aren’t just for conveying information, they’re also designed to entertain.
MetaAI’s conversational interface.
Almost a year after the birth of ChatGPT, Meta Company officially joined the AI chatbot war.
At the annual Connect Developers Conference held on September 27, local time, Meta announced the launch of its own AI assistant MetaAI, and released a series of personalized artificial intelligence characters in its products WhatsApp, Instagram and Messenger.
Although companies like OpenAI and Microsoft started the chatbot race early, given the sheer scale of Meta's social network, its AI assistant may give most people their first taste of generative artificial intelligence.
Provides real-time network results with the ability to generate images via prompts
MetaAI looks a lot like other chatbots of its kind. Meta sees it as a universal assistant that can do everything from planning a trip with friends in a group chat to answering questions that users would typically ask a search engine. Meta announced a partnership with Microsoft's search engine Bing to provide real-time web results, setting it apart from many other chatbots that don't have up-to-date information.
Another important feature of MetaAI is its ability to generate images like those provided by Midjourney or DALL·E through the prompt "/imagine", and the image generation function is completely free.
Ahmad Al-Dahle, vice president of generative artificial intelligence who has been leading the development of Meta's AI assistant, said MetaAI is "based on many of the core principles behind Llama2." Llama2 is Meta's latest open source model and is being rapidly adopted by various industries.
Al-Dal said the rapid popularity of Llama2 has helped Meta improve the way its assistant works. "We just saw a huge demand for the models, and then we saw an incredible amount of innovation happening on the models that really helped us understand their performance, but also their weaknesses, and helped us iterate some of those components and implement them directly into the product."
But MetaAI is different from Llama2, and Al-Dal said his team spent a lot of time "refining additional conversational data sets so that we could create a conversational and friendly tone for the assistant to respond to. A lot of existing AI can be bot-like or bland." Meta expands the model's contextual window "so that we can create deeper and more powerful interactions with the user." MetaAI has also been tuned to give "very concise" answers, he said.
Deep integration with social networks such as Instagram
In addition to Meta's AI assistant, the company has begun rolling out 28 artificial intelligence characters in its apps, some based on celebrities such as Snoop Dogg and Paris Hilton. Others are themed around specific use cases, such as travel agencies.
Meta begins rolling out 28 artificial intelligence characters in its apps.
When a user chats with one of them, their profile image subtly animates based on the conversation.
Character AI dialogue animation.
The new generative AI tool will also allow users to edit images and create memes using text prompts.AI image editing capabilities are rolling out to Instagram, including two new features: restyling and backgrounds. With a redesign, the user enters a text prompt (examples from Meta include "watercolor" or "magazine collage") and the tool updates the existing image based on those instructions.
Restyled functionality.
The Background feature uses the user's text prompts to add a new AI-generated background to an image (e.g. "Surround me with puppies").
Background function.
For both editing features, Meta says it will recognize when an image is created using artificial intelligence tools,So that users can tell whether the content they see is synthetic. The company said it is experimenting with other labeling features, including visible and invisible markers.
AI-generated chat emoticons will be available on Instagram, Facebook, WhatsApp and Messenger. Users can choose to create memes using text prompts, which should "generate multiple unique, high-quality memes in seconds." These features are powered by Llama2.
Users can choose to create emoticons using text prompts
With the rise of TikTok, which has taken away a large number of teenage users from Meta's Instagram in the past few years, attracting young users has always been Meta's top priority.
Currently, MetaAI has not been trained on public user data on Instagram and Facebook, but it may do so in the future.This is a feature that other chatbots cannot copy.. "We see a long roadmap to integrating some of our own social (features) as part of Assistant to make it more useful," Al-Dal said.
Technology companies intensively updated AI functions in September
Meta has long been a powerhouse in artificial intelligence research, powering countless capabilities:From algorithms that display viral content on social networks to systems that flag toxic content.But as Google and OpenAI roll out chatbots and other standalone AI products, Meta seems to be lagging behind in this regard.
This month, tech companies are busy rolling out new experiences and features for generative AI. Google released a new version of its chatbot Bard on September 19, integrating it with several of Google’s most popular services such as Gmail and Docs. On the 20th, Amazon revealed that it will "soon" allow its voice assistant Alexa to conduct "near-human" conversations. On the same day, OpenAI announced the integration of the image generator DALL·E3 into the chat robot ChatGPT. On the 21st, Microsoft announced plans to embed its generative artificial intelligence assistant "Copilot" into many of its products. On the 25th, OpenAI announced the launch of a new version of ChatGPT, which allows users to interact through voice or picture prompts.
Meta CEO Mark Zuckerberg emphasized that Meta's strategy involves creating different AI products for different uses, rather than a single flagship chatbot. Meta's conversational bots aren't just for conveying information, he added, but are also meant to entertain. "I think this is going to change the way people use all of our products," Zuckerberg said.
However, Meta is not the first company to launch an entertainment chatbot.Earlier this year, social app Snapchat launched MyAI, a chatbot designed to provide users with advice on everything from cars to planning weddings. Last year, two former Google employees launched CharacterAI, which allows users to imitate the conversations of Sherlock Holmes or Albert Einstein.
Meghana Dhar, a former Snap and Instagram executive, said: "AI chatbots to me don't exactly fit the characteristics of Generation Z (generally refers to the generation born from 1995 to 2009), but there is no doubt that Generation Z is more familiar with this technology. The younger they are, the more relevant they are to these technologies. The more relatable the bots are. Dahl said that if these AI chatbots can increase the time users spend on Facebook, Instagram and WhatsApp, then they may be beneficial to Meta. “They just want to keep users on the platform longer because it gives them more opportunities to serve ads to users.”
However, researchers have found that giving these chatbots a personality may pose some unexpected challenges. Adding personality to ChatGPT makes its output more toxic, according to a paper published this spring by researchers from Princeton University, the Allen Institute for Artificial Intelligence, and the Georgia Institute of Technology.
U.S. Senator Edward J. Markey, a Democrat from Massachusetts, wrote to Zuckerberg on September 27, urging him to suspend plans to release artificial intelligence chatbots until the company understands the impact they will have on young users. Meta said all AI models have the ability to spread inaccurate or inappropriate content, but the company will continue to improve the safety and accuracy of its chatbots.