Meta puts Llama 3 behind its AI assistant

Meta just dropped its upgraded Meta AI assistant across Facebook, Instagram, WhatsApp, and Messenger, backed by its brand-new Llama 3 model.

Instead of asking billions of people to download another standalone chatbot, Mark Zuckerberg is piping the assistant directly into the search bars and feeds of the apps you already open every single day. Alongside the app rollout, Meta also launched a dedicated web destination at meta.ai, with Zuckerberg boldly claiming it is now likely the most intelligent AI assistant you can freely use.

The global rollout is kicking off in English across more than a dozen countries, including Canada, Australia, New Zealand, Singapore, Pakistan, and South Africa. Interestingly, India was left off the formal launch list despite recent sightings, with Meta confirming it is keeping tests there in a limited, phased capacity for now.

Real-time web search and instant image rendering

Meta is turning its search bar into a direct competitor to traditional search engines. Thanks to live integration with both Google and Bing, you can plan a weekend trip in a Messenger group chat, check flight prices, or find vegan dinner spots with sunset views without ever hopping over to a browser.

The integration goes beyond search bars. While scrolling through your Facebook feed, you can prompt Meta AI directly under a post. If someone shares a photo of the northern lights, you can immediately ask the assistant what month offers the best visibility in Iceland.

Image generation also got a noticeable speed injection through a feature called Imagine. Available in beta on WhatsApp and the web in the US, the tool updates images in real time as you type each letter. The model renders text in images much more legibly than before and lets you animate still results into GIFs with a single prompt. Meta AI is already live on Ray-Ban smart glasses and is heading to Quest headsets next.

Hallucinations inside private group chats

Shoving an AI assistant into every corner of a social network comes with massive moderation headaches. Large language models still hallucinate, and in social contexts, that can get weird fast.

Earlier this week, 404 Media reported that Meta AI chimed in on a real Facebook parents group, claiming it had a gifted child attending a specific public school in New York. Befuddled parents flagged the bizarre intrusion, forcing Meta to step in, delete the response, and remind users that the system is still learning.

Putting an eager bot inside sensitive community spaces requires strict boundaries, something Meta is still fine-tuning on the fly.

Why it matters

We think this is the most aggressive distribution play the consumer AI space has seen yet. While OpenAI and Anthropic are battling for paying subscribers and enterprise seats, Meta is giving away a frontier-grade model to over three billion daily active users for free. For average people who never cared to bookmark ChatGPT, having an assistant embedded into their existing WhatsApp or Instagram workflow means they might never need to look elsewhere.

It also puts immense pressure on Google. By utilizing both Google and Bing under the hood to answer conversational search queries inside social messaging, Meta is capturing search intent at the source. Why leave Instagram to search for a restaurant when the app’s search box handles it natively?

However, turning social feeds into AI playgrounds carries real risk. The recent incident of Meta AI fabricating a personal life in a parents forum proves that social AI needs fundamentally different guardrails than an enterprise copilot. If Meta cannot stop its assistant from disrupting genuine human connection with nonsensical interventions, user fatigue could set in quickly. Still, as a pure land grab for mainstream AI usage, Meta's move is brilliantly executed.