Chile unveils an open-source AI model trained on regional data to combat US-centric bias and promote cultural sovereignty in technology.
SANTIAGO: Chile has launched an open-source artificial intelligence model designed to combat the US-centric bias prevalent in the global industry.
The model, named Latam-GPT, was developed by the Chilean National Center for Artificial Intelligence (CENIA) using millions of data points collected across Latin America.
President Gabriel Boric said the initiative positions the region as an active player in the future economy. “We’re at the table — we’re not on the menu,” he added.
Science Minister Aldo Valle stated the program was built to combat prejudices and generalizations about the region. He warned that being a passive user of AI could result in the loss of significant cultural traditions.
Unlike closed models like ChatGPT, Latam-GPT is open for programmers to customise. CENIA director Alvaro Soto noted that global models contain very little Latin American data.
This lack of representation is sometimes reflected in AI depictions, such as ChatGPT portraying a typical Chilean man in a poncho with the Andes in the background.
The model was trained on over eight terabytes of data, equivalent to millions of books, for a cost of just $550,000. Funding came primarily from the Development Bank of Latin America and CENIA’s own resources.
Contributions to the project were provided by universities, foundations, and government entities across eight Latin American countries. It is currently trained mainly in Spanish and Portuguese content.
Developers plan to incorporate material in Indigenous Latin American languages in future iterations. The model will be available free of charge to companies and public institutions.
Soto cited potential applications for hospitals with logistical or resource management problems. Professor Alejandro Barros said the model has “no chance” of competing directly with major AI giants due to its tiny budget.
However, it has already attracted interest from local entrepreneurs like Roberto Musso. His company Digevo plans to use Latam-GPT to develop customer service programs for airlines and retailers.
Musso said clients want users to express themselves and receive responses in the local language. He highlighted the model’s ability to recognise regional slang, idioms, and speech patterns.
A first version was developed on Amazon Web Services, but future training will occur on a supercomputer at the University of Tarapaca. The launch follows similar regional AI projects in Singapore and Kenya.
