An almost unprecedented large-scale language model, Meta AI, a virtual assistant designed to generate photorealistic images and provide detailed answers to user queries, is designed to deliver “more detailed answers on mobile. ” and a more accurate overview of your search results, making it more convenient for you,” said Social Media. The Giants announced today.
In a blog post, Meta revealed more than 20 new AI-driven features that will be added to the app to upgrade everything from search functionality to social media experiences to business communications.
“We are evolving Meta AI to add new capabilities to the messaging experience, while also leveraging it behind the scenes to power smart capabilities,” said Meta.
The company also leverages the power of EMU (Meta’s native text-to-image model) to create a standalone image product that allows users to easily create a variety of images, similar to DALL-E 3 and Stable Diffusion. We are launching Imagine as a generator. This functionality was originally designed to run within a chat application, but was split into a web-based tool to compete more directly with the industry’s top models.
For its built-in marketplace, Meta also offers vendors and sellers the option to leverage generative AI to improve the user experience. The idea is to “give people in different English markets the option of AI-generated post comment suggestions, community chat topic suggestions in groups, search results, and in-store product copying.” ”, the company said.
However, according to Meta’s official post published in Portuguese, Imagine with Meta and other generative AI experiences are not available in Brazil, Mexico, and Argentina.
To promote transparency in AI-generated content, Meta said it will implement an “invisible watermark” on images generated by its AI tools. Although watermarks are not a particularly strong way to identify AI-generated content, this move should help address concerns about the potential misuse of AI in content generation.
These new AI capabilities have already been well received. “I’m no expert, but the generation seems good,” says Julien Chaumond, co-founder of Huggingface, the world’s largest community for open source AI. “Now he’s 2023, but I was do not have I don’t think I’ll ever have to log into Facebook on a computer again. ”
In the announcement, Meta highlighted the continued rollout of AI experiences across applications, the release of the Llama family of large-scale language models, and research advances such as Emu Video and Emu Edit. These will continue to enhance the hardware capabilities needed to power Meta’s products next year, especially AR and VR applications, the company said.
Meta AI is also being leveraged to enhance the user experience on Facebook and Instagram. For content creators, Meta is using the LlaMA text generator to test AI-generated reply suggestions within direct messages. This feature is designed to facilitate more efficient communication between creators and viewers by crafting replies that match the creator’s communication style.
AI is also involved in creating personalized greetings, editing posts, and providing content suggestions in different sections of the platform, such as Groups and Marketplace.
Meta says its AI models have long-term memory, which makes them even more functional and useful, and solves the main problem of long-duration interactions with AI assistants and characters that rely on LLM: He said that the longer you go, the more you can deal with the problem of losing track of information.
Meta also plans to incorporate Multi-Round Automatic Red Teaming (MART) into its AI models. The technology essentially trains adversarial models to help Meta detect offensive, deceptive, or harmful content by pitting the models against each other while both evolve and become more capable. To do.
as Decryption As previously reported, Meta recently disbanded its “Responsible AI” team, choosing instead to distribute its tasks among various departments involved in AI development. Meta’s AI models have already been trained on billions of social media posts.
Edited by Ryan Ozawa.