Scholar Helps Build a ChatGPT for Latin America

Large language models like ChatGPT are only as useful as the data on which they are trained. That is why top technology innovators, like Chile’s National Center for Artificial Intelligence (CENIA), value the expertise that scholars like Freddy Vilches, associate professor and section head of Hispanic studies, bring to the development of generative AI projects.

“You don’t want the models trained on superficial things, but rather on information that provides an in-depth look into a given subject,” says Vilches who, in addition to his faculty position, is also an accomplished composer and musician. “You want material that is legitimate.”

Vilches is working with the center on a number of projects to advance its mission to “promote an ethical and humane technological future.”

Freddy Vilches, associate professor and section head of Hispanic studies, is also an instructor of Charango and Venezuelan Cuatro. His AI collaborators value his background in the arts and humanities.
Credit: Nina Johnson

Creating LatAmGPT

Three years ago, motivated by a desire to include diverse perspectives, CENIA invited Vilches to contribute an article to an index they were compiling on the status of AI throughout Latin America. The following year, he served as coeditor on the follow-up edition of the index. When the group started work on LatAmGPT, a large language model tailored to Latin American and Caribbean interests, they welcomed Vilches as a collaborator. He joined contributors from more than 30 institutions from countries around the world.

“I think what they value about my contribution is the fact that I come from the humanities and the arts,” says Vilches. “I’m from Chile, like many of my colleagues at the center, and I’ve also spent a lot of time in the United States, as well as in Mexico, Cuba, and other Latin American countries because of my research in literature and in music. So I have an outlook on Latin American culture from a variety of countries and viewpoints.”

In addition to providing articles and books for the model to train on, Vilches is also working with the team to curate the data and decide how the system will pick and choose information in response to user prompts.

“We want it to be as Latin American as possible, to reflect the diversity of our culture,” says Vilches. “There are elements of Latin America in Oregon, in California, in Texas, and other states.”

Vilches sees LatAmGPT as a complement to the more general information that ChatGPT might provide. “LatAmGPT will be more curated to reflect the depth of knowledge in Latin America,” he says.

Preserving Indigenous Languages With AI

Researching Indigenous languages has long been a part of Vilches’ work as a scholar and composer. For example, he worked closely with different communities in Mexico, Bolivia, Chile, Guatemala, and Peru on the creation of a choral composition that incorporates the languages of the Aymara, Quechua, Mapuche, Mayan, and Nahuatl peoples.

Now Vilches is working with CENIA and several indigenous communities on an AI-based translator for Indigenous languages. A top priority for Vilches and his center collaborators is to make sure the translation AI is truly a collaboration with the Indigenous communities who speak those languages. The goal is for the communities to take ownership of these projects.

“Part of the effort is to help preserve those languages,” says Vilches, “and part of the plan is to develop learning tools for the community members to use.” These learning tools will be designed by each community to preserve and teach these languages to younger generations.

The team started with two languages: Mapudungun, spoken by the Mapuche people in parts of southern Chile and Argentina, and Rapa Nui, the language of the people of Easter Island, part of Chile.

“In the case of Rapa Nui, CENIA has worked with local linguists, as well as community elders, to document how the language has changed,” says Vilches. “The process has been really respectful of those communities. We want to make a contribution by helping with the technical side so it can be of use to them in promoting, preserving, and teaching the language.”

Vilches shares this work with students in his L&C classes. In spring 2025, one of his collaborators from Chile visited an upper-division Spanish language class to talk to students about the AI projects.

“I think students really appreciated learning about what I’m doing that’s related to literature and Latin American studies outside the college,” says Vilches. “They particularly like hearing first hand about this group based in Chile, which is doing important cultural and technological work and is also a leader in terms of AI initiatives in Latin America.”

Hispanic Studies