Meta Expands AI Assistant to 22 Countries, Mexico Included

Meta has expanded the availability of its artificial intelligence assistant, Meta AI, to 22 countries, including Mexico, Argentina, Colombia, Chile, Ecuador and Peru. This expansion allows Facebook, Instagram, and WhatsApp users to interact with the assistant in multiple languages, including Spanish, English, German, French, Italian, and Portuguese.

On Facebook and Instagram, the assistant can be found in the search bar, direct messages, and the news feed, while on WhatsApp, Meta AI has its own conversation window and can be used in individual and group chats via the @Meta AI command. Meta AI not only answers general questions but can also generate images and provide real-time information based on search engines such as Google and Bing.

The assistant's ability to generate images from text prompts and answer questions related to content posted on the platforms adds an additional level of interactivity and personalization for users, the company notes. 

Meta AI will also soon be integrated into Meta Quest, the company's line of virtual reality (VR) devices, in the U.S. and Canada, initially in experimental mode, replacing current voice commands to enable hands-free control of the device. 

Users will be able to interact with the assistant to get real-time answers, check the weather and more, all while using Meta Quest in mixed reality mode. In addition, with the Vision in Passthrough feature, Meta AI will offer contextual assistance, such as weather-based outfit recommendations or suggestions for complementary outfits, and provide local information, such as restaurant recommendations. 

The 'Imagine Me' feature, expected to be integrated in all company markets but currently in beta and available only in the United States, allows users to create digital representations of themselves in various environments, which adds a component of creativity and personalization to Meta AI's offering, the company said.

Meta has highlighted that its AI assistant uses the Llama 3.1 model, its largest and most advanced open source AI model to date, with 405 billion parameters. This version, according to information from The Verge, outperforms competitors such as GPT-4o and Anthropic's Claude 3.5 Sonnet by several benchmarks. 

Trained with the more than 16,000 graphics processing unit (GPU) of Nvidia's H100, Llama 3.1 can perform complex reasoning and handle advanced technical data. 

Meta is collaborating with key partners such as Microsoft, Amazon, and Google to integrate Llama 3.1 into various applications, highlighting its lower cost compared to private models such as GPT-4o. The company is also testing Llama 3.1 in cybersecurity and biochemistry scenarios and has incorporated new functionalities, such as the ability to generate images based on the user's visual context. 

Meta has acknowledged concerns about the accuracy and appropriateness of the assistant's responses. The company updates the system every two weeks to improve its performance and user experience.

The integration of AI features has also raised concerns about user privacy. Meta has admitted that it used 1.1 billion images from Facebook and Instagram to train its AI-based image generation engines, and has adjusted its terms of use to allow the use of old user posts as data to optimize its AI systems.

The recent modification of its terms of use to include old user posts as training data has raised concerns among data protection authorities, Wired notes. Meta, which faces investigations in the European Union and the United States, defends its approach by claiming that openness in AI development promotes safer and more innovative products, although these practices continue to face scrutiny for their impact on privacy.