Following the launch of Llama 3–Meta’s newest ‘most powerful’ large language model–Meta has announced that its AI chatbot, MetaAI (which is now powered by Llama 3), will now be accessible on its WhatsApp, Facebook, Instagram, and Messenger apps, across the following 12 new countries:
- Australia
- Canada
- Ghana
- Jamaica
- Malawi
- New Zealand
- Nigeria
- Pakistan
- Singapore
- South Africa
- Uganda
- Zambia
- Zimbabwe
Plus, it’s already available on Ray-Ban Meta smart glasses in the US and will be integrated into the Meta Quest VR headset over the coming months.
Users can either access MetaAI via the search bar, within one-on-one or group chats, or on their feed, on all four platforms. Or, they can access it via a new, standalone site: meta.ai, where they can chat with the model, and save and view their conversations, like they can with ChatGPT and Google’s Gemini.
Users can ask MetaAI questions and it will surface relevant information from reels, posts, chats, and messages, as well as real-time search engine result summaries, as it’s powered by Google and Bing. For example, users could ask it for restaurant recommendations, concert listings, recipe suggestions, holiday itineraries, or help with exam studies.
MetaAI also has a new image generation feature, called “Imagine” (currently in beta) which can create an image, in real-time, as a user is typing. It can also animate the image or turn it into a GIF.
Llama 3 is reportedly smarter than its predecessor, with improved reasoning skills and a better ability to follow instructions, meaning MetaAI is capable of doing more human-like tasks than ever before.
Meta has also taken safety seriously with Llama 3, stating that they had stress-tested the model, refining, and training it not to discuss illicit or inflammatory topics relating to things like weapons or cyber-attacks. However, it has provided a disclaimer to this, expressing that “generative AI doesn't always get it right and could further spread misinformation or entrench biases.”
They’ve also openly declared that the model has been fine-tuned on information from its other services, along with data from user interactions with Meta AI. So, this means conversations with Meta AI might also be used to train the model in the future, with no apparent way to opt out, like you can with ChatGPT.