Artificial intelligence will flood the networks: Meta launches its model on WhatsApp, Facebook and Instagram | Technology

Meta announced this Thursday its great landing in the competitive artificial intelligence (AI) sector. Users of all its applications, some of the most popular on the planet, will have an assistant to whom they can ask questions, ask for advice or ask for drawings. WhatsApp, Instagram, Facebook and Messenger will encourage from a prominent place to ask questions to Meta AI. Meta will also create a separate app and page to consult your assistant for free, in the same way that you can now do with ChatGPT or Google’s Gemini.

Movement is Meta’s big leap in AI. This Thursday it also presented Llama 3, the new version of its model that will boost all its attendees. Meta’s decision will popularize access to AI in an extraordinary way. Its apps are used by more than 3 billion people around the world. Although the success of OpenAI’s ChatGPT has been dazzling, this new level of access is significant. 11% of Spaniards say they use ChatGPT at least once a day.

The assistant will not only accept questions and queries, but also create illustrations and animate them in real time, to make something similar to a GIF file, as seen in the company’s demos. It will work in a similar way to the rest that already exists on the market, with the notable novelty that the creation of images is immediate: while the user writes, the machine adds or removes details from the drawing.

The launch of Meta IA will be available, for now, only in English and in 14 countries, such as the United States, Canada, Australia, Nigeria, Pakistan or South Africa, among others. No European or Latin American country, for now. The company says it has “nothing to share” about the arrival of these news in Spanish or to Spanish-speaking countries when asked by EL PAÍS. In his presentation post, Meta explains that “the models we launched today are only refined for results in English.”

The company, however, already warns that its results in other languages ​​they will be worse: “To prepare for the upcoming multilingual use cases, more than 5% of the Llama 3 pre-training data set consists of high-quality data in languages ​​other than English, covering more than 30 languages. However, we do not expect the same level of performance in these languages ​​as in English.”

It is difficult to predict how much and how people will use these in-app virtual assistants. In September, Meta launched 28 AI chatbots in the US with profiles such as Kendall Jenner, Paris Hilton or rapper Snoop Dogg. They have not been a success. Snoop Dogg has just 14,000 subscribers and his Instagram account has 86 million. But the new arrival of AI can provide new attractions for users.

And like the rest of the previous releases in this field, Llama 3 will not be exempt from the errors and hallucinations that its competitors have been suffering from. There are those who have already found a black Italian fascist soldier, as happened a few weeks ago when Google’s AI drew some black Nazis. Given that these tools will be available on a massive scale, in all networks and on a planetary scale, we are already alerted to the challenge of misinformation and infoxication that they may end up generating.

Another novel capability of the virtual assistant is that it has the ability to insert itself into group conversations, apparently to encourage them. This week, a Princeton University professor posted a screenshot of a private Facebook parenting group where Meta AI claims to have a child in a New York public school and shares her child’s experience with teachers.

The Meta algorithm ranks it as the top comment. A Facebook help page says that Meta AI will join a group conversation whether it’s invited or if someone “asks a question in a post and no one responds within an hour.” This makes it evident that, beyond the use that users will give to the bot, a clear objective is to promote attention and increase user interactions with each app.

Meta has already had to apologize at least a couple of times for comments or responses from his bot. In both cases she has given an excuse linked to the known errors of these bots: “This is new technology and will not always return the response we expect, which is the same thing that happens to all generative AI systems,” declared the company to the AP agency.

These errors and outbursts will be inevitable. Meta has refined his model to be less shy when answering questions about sensitive topics. “If someone asks about a controversial political issue, our goal is for Meta AI to not offer a single opinion, but rather to summarize relevant views on the topic,” says the company. But, she continues, “if someone asks about a view on an issue, we want to respect that person’s intent and have Meta AI answer the specific question.”

In addition to expanding the reach of AI to hundreds of millions of people, Meta is also trying to push into the domains of OpenAI, Google and Microsoft with its free assistant page, Meta.ai. The company continues to train a better version of Llama, according to Zuckerberg. If it remains free, it may give Meta a competitive advantage over the paid versions of OpenAI (ChatGPT 4) and perhaps Google (that would be studying the possibility of launching a premium search engine).

In the presentation of this release on his Instagram account, Mark Zuckerberg, executive president of Meta, appeared with an appearance and chain that was highly commented on on the networks. Zuckerberg said Llama 3 would be “the smartest AI assistant that’s free to use.” All the big corporations and the start-ups of the sector (the French Mistral, Anthropic or Perplexity) try to convince users that their assistants or models are the ones that best resolve any issue and in the most agile way. For now, it remains difficult to distinguish the unique traits of each attendee and there is no standardized system that allows for rigorous comparisons.

The company has not wanted to explain with transparency what type of content it has trained its model with. The shortage of new material to feed these programs is one of the big debates within the sector and Meta has admitted that it has used AI-generated language (also called “synthetic data”) to train. Until now, experiments with non-human language to expand the amount of data have been unsuccessful: “We also use synthetic data to train in areas such as programming, reasoning and large contexts. For example, we use synthetic data to create longer documents to train on,” the note explains. Goal presentation.

You can follow The USA Print in Facebook and x or sign up here to receive our weekly newsletter.