Latam-GPT, the AI ​​that seeks to break prejudices about Latin America

The project is promoted by the National Center for Artificial Intelligence of Chile (Cenia), a private corporation with public financing.

LOOK: The success of “disconnection” applications, a wall against algorithms

Latam-GPT has the support of universities, foundations, libraries, government entities and civil society organizations from countries such as Chile, Uruguay, Brazil, Colombia, Mexico, Peru, Ecuador and Argentina.

Latam-GPT wants to break “prejudices” and prevent the representation of Latin America in the rest of the world from “looking all the same,” Chilean Science Minister Aldo Valle tells AFP.

The region “cannot just be a passive user or recipient of artificial intelligence systems. This may result in losing a good part of our traditions,” he adds.

Despite its name, this tool is not a chat that you can interact with. It is a “large database” trained from information from the region, which can be used to develop technological applications, explains Minister Valle.

The development of large AI models is mainly concentrated in the United States, China and Europe. In addition to Latam-GPT, other alternatives have emerged, such as SEA-LION in Southeast Asia or UlizaLlama in Africa, more focused on their own cultures.

To train Latam-GPT, more than eight terabytes of information were gathered, equivalent to millions of books.

This AI was created with just $550,000 in financing, coming mainly from the Development Bank of Latin America (CAF) and its own resources, in addition to agreements with other institutions.

Its first version was developed in the Amazon Web Services cloud, but in the future it will be trained on a supercomputer that must be assembled at the University of Tarapacá in the north of the country and that cost almost 5 million dollars.

“The models developed in other parts of the world, although they have data from Latin America, are a fairly small proportion,” Álvaro Soto, director of Cenia, tells AFP.

At this stage, the model information is mainly in Spanish and Portuguese, although the objective is to also incorporate content in indigenous languages ​​of the continent.

However, “there is no possibility that (Latam-GPT) can compete against the large AI models,” Alejandro Barros, professor at the Department of Industrial Engineering at the University of Chile, tells AFP.

For the academic, the budget difference is key, since other projects have “hundreds of millions of dollars and in some cases billions of dollars for infrastructure.”

Latam-GPT is a free platform that should be used to develop applications and technology “more specific to the region,” says Soto.

An example of this could be digital tools for “hospitals with logistical problems or the use of medical resources.”

One of the first companies to use Latam-GPT will be the Chilean Digevo, which will develop conversational robots specialized in customer service for airlines or commerce companies.

These companies “are very interested in their users expressing themselves and receiving responses in local language,” Roberto Musso, director of Digevo, a company dedicated to developing digital applications, told AFP.

According to Musso, Latam-GPT provides the ability to recognize “jargons, idioms and even the speed of speaking” and avoid “bias problems” that can arise in other AI models.