The technology giant Meta (formerly Facebook) does not stop there - tonight (Thursday) it revealed a series of innovations in the field of artificial intelligence, which include a personal artificial intelligence assistant Meta AI that will be integrated into the company's applications; A new open source language model that records industry-leading performance (Meta Llama 3) that was also used to build the Meta AI assistant; A new artificial intelligence tool that turns text into images in real time and can also animate them - and will be available in a beta version in the United States (Imagine), as well as new safety measures that will ensure open and responsible innovation.
The Meta AI personal assistant is coming to apps, and is gradually expanding to additional countries
The Meta company announces tonight that the personal artificial intelligence assistant, Meta AI, will gradually become available worldwide through the Facebook, Instagram, WhatsApp and Messenger apps, as well as in the Ray-Ban Meta and Meta Quest smart glasses. The AI assistant launches tonight, in English, in more than 12 countries outside the US (including Australia, Canada, Ghana, India, Jamaica, Malawi, New Zealand, Nigeria, Pakistan, Singapore, South Africa, Uganda, Zambia and Zimbabwe) - and this is just the beginning.
The Meta AI assistant is developed on the basis of the new Meta Llama 3 language model, it is already considered one of the leading artificial intelligence assistants in the world in terms of performance, and will be gradually available worldwide. In addition, the Meta AI assistant will also be accessible via the computer browser. At meta.ai, Meta's artificial intelligence assistant will be able to perform a variety of actions - from searching for restaurants, through recommendations for vacations and attractions, to helping with studies, solving mathematical problems or getting inspiration for the design of the new apartment. The
Meta AI assistant will also be available in the search on Facebook, Instagram, WhatsApp and Messenger - and people will be able to access information in real time from around the web, without closing the apps and switching to another app. Also, people will be able to access Meta AI when they scroll through their Facebook feed. If they come across a post that interests them - they will be able to request From Meta AI more information on the topic, directly through the post itself.
New open source language model: Meta Llama 3
Meta Company releases the first two models of Meta Llama 3, which are now available for widespread use. This early release includes pre-trained language models with eight billion and 70 billion parameters, which can support a wide variety of use cases. The new generation of Llama records industry-leading performance in a wide range of measurements compared to open models, and is comparable to other commercial models.
The first models will be released in open source, for the benefit of the developer community, as part of Meta's long-standing commitment to open innovation in artificial intelligence. Thus, the company adopts the open source ethos of early release, allowing the community to gain access to models while they are still in development. The text-based models released today are the first in the Llama 3 collection, with the goal being to make Meta Llama 3 a multi-lingual and multi-modal model, and continue to upgrade the model with a variety of capabilities and tasks (such as reasoning or writing code).
A new AI tool for creating real-time images: Imagine
In addition, the Meta company is presenting tonight a new tool for creating images from text commands in real time. The tool will be launched in beta in the United States, on WhatsApp and on the web version of Meta AI.
When people give the Imagine command in a chat with Meta AI, they will be able to formulate a prompt - and the image will start to appear as they type. The generated image will change as each additional letter is typed, so people can see in real-time how the AI generates the image and brings their vision to life. The technology giant is also introducing new capabilities to animate photos with the help of the artificial intelligence assistant, which will be able to turn photos into animations and even into GIF files.
New safety measures to develop responsible innovation
Meta Company is committed to developing artificial intelligence responsibly, and helping others do the same. The company is sharing tonight how it refined the way the personal assistant Meta AI will respond to prompts on political or social issues, and incorporated guidelines to ensure that the AI assistant does not present a single opinion or point of view, but summarizes various relevant perspectives on the subject. Meta also used red teams and external experts to identify unexpected ways people might abuse Meta AI or the Llama 3 model. The company also implemented safeguards - both at the prompt level, and at the output level of the AI assistant.
In addition, Meta expanded the training set for Llama 3 so that over 5% of the model's early training dataset consists of high-quality non-English data, covering more than 30 languages. Also, Llama 3 is trained on a variety of public data, and for the purpose of training Meta removed data from sources known to contain a large amount of personal information. During and after model training, Meta performed automated and manual assessments—along with red teams and external experts—to understand the models' performance across a range of risk domains.
Over the coming months, Meta will release additional Llama 3 models with new, multimodal capabilities, including multi-language conversation, and stronger overall performance. Meta is currently training a model as large as 400 billion parameters - and any final decision about its open source release will be made following the safety assessment that will take place in the coming months.