The demise of Facebook has long been predicted, but just like NewstalkZB, it actually gets more entrenched as its audience ages. Facebook was used by 79% of New Zealanders aged 16-64 last year, a rate of usage among the highest in the Western world.
You may have noticed the curious iridescent circle that has appeared at the top of the Facebook app on your smartphone, as well as on Instagram and Meta’s messaging apps, Messenger and WhatsApp. Occasionally the circle contorts itself into the shape of the Meta logo, the symbol for infinity.
Welcome to Meta AI, the new artificial intelligence assistant that has been embedded into Meta’s social media apps. The tag “Ask Meta AI” now appears beneath some posts, too, followed by automatically generated questions designed to help you find out more about the subject of the post.
When you ask Meta AI a question, or tap on one of the prepopulated ones, you end up in a version of Facebook Messenger, conversing with Meta AI. You can ask it to generate images, too. The results are of varying quality, but similar to other AI chatbots you may have used. Use it to get quick factoids or recipe suggestions. Don’t rely on it for medical advice. Meta itself admits its AI bot is subject to bias and inaccuracies.
Meta’s AI-generated suggestions at the top of the app, supposedly based on things it thinks I’m interested in, also need some work. “Pilates reels”, “small garden ideas”, and “imagine a cat cafe on Mars” are just not the sort of conversation prompts I’m ever going to tap on.
Eventually, the chatbot will direct you to the Google search engine if you’re looking for specific information. The ChatGPT-like bot is underpinned by Meta’s Llama 3 large language model, which has scraped vast amounts of text and imagery from the web to train it to give convincing answers.
Unlike ChatGPT and other large language models from the likes of Google, ChatGPT creator OpenAI and Anthropic, Llama 3 is an open-source model. That’s a good thing and a surprising move by Meta, considered the rapacious Galactic Empire of the social media world. It means anyone can freely use it to create their own AI chatbots. Local researchers are already exploring how it could be applied to scientific endeavours.
Meta AI is not as sophisticated as the king of the bots, ChatGPT. It has up to 70 billion parameters, according to Meta boss Mark Zuckerberg, compared with a rumoured 1.8 trillion for OpenAI’s GPT-4. Generally, the more parameters in a large language model, the better the output will be.
But OpenAI charges US$20 a month for access to its most sophisticated AI. Meta AI is free to us, and now on the devices of billions of users, giving the company a big advantage over its rivals.
Where is this taking social media? We’ve known for years that Facebook has been harvesting our data to sell targeted advertising. AI bots represent the motherlode for Meta: the questions we ask will serve to fine tune the algorithms, giving it that much more intel on us.
The company says it captures chat conversations but makes them anonymous, so doesn’t connect them to your identity. But other data points can be used to link a user with chat conversations. Either way, it has a rich new flood of data to mine for advertising and to increase engagement with its apps.
By the way, you can’t turn Meta AI off. Replies can be muted, but the AI bot will still appear when you search across Meta’s apps. Once again, when the price to access is “free”, you are the product being sold.
