Chile’s National Center for Artificial Intelligence (Cenia) saw the need to promote the creation of Latam-GPT. This will be the first collaborative artificial intelligence model in Latin America and the Caribbean.
What is Latin American culture? For ChatGPT, it’s a vibrant mix of indigenous roots, African influences, and European heritage. This is expressed through music, art, gastronomy, and the warmth of its people.
Latam-GPT: The AI that Speaks Our Language and Understands Our Culture
Announced at the AI for Good Summit in Paris, Latam-GPT promises to be a large language model, similar to ChatGPT or DeepSeek, but with a crucial focus: reflecting the culture, language, and history of the region.
This will allow it to offer more accurate and representative information about local contexts and address a need that global models don’t always meet.
This project has been under development for two years with the support of the Chilean Ministry of Science, Technology, Knowledge, and Innovation. It’s a collaborative effort that brings together more than 30 institutions and 60 experts from across the region.
Countries like Mexico, Colombia, Ecuador, Argentina, Peru, Uruguay, Costa Rica, Spain, and the United States have joined the initiative, contributing data, knowledge, and technical expertise. The result is a data corpus that already reaches 50 billion parameters, comparable to OpenAI’s ChatGPT-3.5.
Relevance of Latam-GPT
“The importance of this model lies in its ability to embody the culture, idiosyncrasy, and worldview of Latin America in a transformative technology,” explains Rodrigo Durán, manager of Cenia Chile. “This collaborative effort, with contributions from various countries, ensures that the solution reflects the richness and diversity of our region,” he adds.
Latam-GPT not only demonstrates that Latin America can develop cutting-edge artificial intelligence, but also puts it at the service of its inhabitants, proposing specific solutions for the regional context. To feed its algorithm, more than 8 terabytes of information have been collected from public and private virtual libraries. Chile, the project coordinator, will have the high-performance infrastructure of the University of Tarapacá (UTA).
This institution has invested in a supercomputer that will enable the training of the language model, a key piece for the development of Latam-GPT. The processor, currently in the import process, will be fundamental for training the model and its subsequent launch, scheduled for mid-June 2025.
With Latam-GPT, artificial intelligence will speak our language and understand our culture.
Latam-GPT: Technological Sovereignty or a New Form of Digital Colonialism?
The promise of Latam-GPT, the first collaborative artificial intelligence (AI) model in Latin America and the Caribbean, is attractive: an AI that understands our culture, our language, and our history. However, Ulises A. Mejías, an expert at the intersection of technology and power, invites us to look beyond the initial enthusiasm and question the fundamental premises of this type of project.
Mejías, who holds a Doctorate of Education (EdD) in Communications, Computing, and Technology in Education from Columbia University and is a professor at the State University of New York, argues that the datafication of our lives represents a new form of colonialism. From this perspective, the development of Latam-GPT, although ambitious and well-funded, does not guarantee greater technological sovereignty for the region.
“I don’t trust projects that try to differentiate themselves from the generative AI (GenAI) models offered by companies in the US and China, but don’t question the basic premise of these models,” Mejías states in an interview with BBC Mundo.
The expert questions whether Latam-GPT truly offers a new vision on the utility of generative AI, or simply reproduces the logic of reducing labor costs and maximizing corporate profits.
According to Mejías, the proliferation of AI models “for specific regions or minority groups” doesn’t address the central problem: What is the purpose of GenAI and who benefits? “Is Latam-GPT attempting to provide a new answer to this question?” he adds.
Environmental Impact: An Unavoidable Cost?
Training large language models has a considerable environmental impact, in terms of carbon footprint, energy consumption, and water footprint.
While Cenia Chile assures that Latam-GPT’s training will have a lower impact thanks to the abundant renewable energy in the north of the country (estimating emissions of 0.96 tons of CO2 and energy consumption of 135 kWh in its first stage), these are preliminary estimates.
The true environmental impact can only be confirmed after the model’s training. The key question is whether the promise of a “local” AI justifies the environmental cost, considering global concerns about climate change.
Data Protection: A Pending Challenge
Another crucial point is the handling and protection of data, both public and private, that will feed Latam-GPT.
While the project’s promoters assure that policies exist to safeguard intellectual property, questions remain about privacy and access to information. In a context of growing concern for data security, transparency and accountability are fundamental.
Latam-GPT presents a dilemma: Does it represent a step towards technological sovereignty, or simply an adaptation to a pre-existing digital paradigm with its inherent problems of data colonialism and environmental impact?
The debate is open and the answer remains to be written.
