Latam-GPT is a Chilean-driven artificial intelligence model to train applications with Latin American data, reduce bias and provide a more accurate representation of the region in a sector dominated by US developments.
Chile has presented Latam-GPT, a project aimed at providing Latin America with its own artificial intelligence (AI) model in a sector dominated by American companies and with the objective of limiting the biases observed in current systems.
The initiative has been promoted by Chile's National Centre for Artificial Intelligence (Cenia), a private corporation with public funding. It has also been supported by universities, foundations, libraries, governmental entities and civil society organisations in countries such as Chile, Uruguay, Brazil, Colombia, Mexico, Peru, Ecuador and Argentina.
During the presentation of the project on Television Nacional this week, the Chilean President, Gabriel Boric, highlighted the region's role in global technological development and welcomed the fact that Latam-GPT will enable Latin America to position itself as an active player in the economy of the future.
The Chilean Minister of Science, Aldo Valle, said that the tool seeks to break down prejudices and prevent the representation of Latin America in the world from being homogeneous. According to Valle, the region cannot just be a passive user or recipient of artificial intelligence systems developed elsewhere, as this could result in the loss of part of its traditions.
Despite its name, Latam-GPT is not an interactive chat. It is a large database trained on information from the region, designed to serve as a basis for the development of technological applications.
A development with regional scope
The development of large artificial intelligence models has been mainly concentrated in the United States, China and Europe. In addition to Latam-GPT, other regional initiatives have emerged, such as SEA-LION in Southeast Asia or UlizaLlama in Africa, focusing on their own cultural contexts.
To train Latam-GPT, more than eight terabytes of information have been collected, a volume equivalent to millions of books. The model has been created with funding of $550,000 (more than €460,000), mainly from the Development Bank of Latin America (CAF), in addition to its own resources and agreements with other institutions.
The first version of the system has been developed in the Amazon Web Services cloud. In the future, the model will be trained on a supercomputer to be installed at the University of Tarapacá, in northern Chile, during the first half of 2026, with an investment of close to five million dollars.
Alvaro Soto, director of Cenia, pointed out that the models developed in other regions of the world include Latin American data in a reduced proportion.During the launch, President Boric exemplified this difference by comparing the extensive information available on the siege of Calais with the much more limited coverage of key battles of Chilean independence, such as the siege of Chillán.
At this stage, the content of the model is mainly in Spanish and Portuguese, although the aim is to incorporate indigenous languages as well.
However, academic Alejandro Barros of the University of Chile has warned that Latam-GPT has no chance of competing with the large global artificial intelligence models because of the difference in economic resources and infrastructure.
Possible uses
Latam-GPT will be a free platform and could be used to develop tools adapted to local needs, such as digital solutions for hospitals with logistical problems or the use of medical resources, according to Soto.
One of the first companies to use the model will be Chile's Digevo, which will develop conversational robots specialising in customer service for airlines and retail companies. According to its director, Roberto Musso, these companies have shown interest in having users express themselves and receive answers in the local language.
Musso said Latam-GPT can recognise slang, idioms and even the speed of speech, reducing the problems of bias that can occur in other artificial intelligence models.
