Latam-gpt is modern A enormous language model developed in Latin America and Latin America. The project, directed by non -Profit Chilean National Center for Artificial Intelligence (Calia), aims to lend a hand the region achieve technological independence by developing the AI Open Source model trained in Latin America’s languages and contexts.
“This work cannot be undertaken by one group or one country in Latin America: this is a challenge that requires everyone’s participation,” says álvaro soto, director of Czeń, in an interview with Wired en Español. “Latam-GPT is a project that aims to create an open, free, and above all AI model. We have been working with a very bottom-up process for two years, combining citizens from various countries who want to cooperate. Recently he has also seen top-down initiatives, and governments are taking interest and are starting to participate in the project.”
The project is distinguished by its spirit of cooperation. “We do not want to compete with OpenAI, Deepseek or Google. We want a model specific to Latin America and the Caribbean, aware of cultural requirements and challenges, which include, such as understanding of different dialects, the history of the region and unique cultural aspects,” explains Soto.
Thanks to 33 strategic partnerships with institutions in Latin America and the Caribbean, the project gathered a corps of data exceeding eight terabytes of text, equivalent to millions of books. This information base has enabled the development of a language model with 50 billion parameters, a scale that makes it comparable to GPT-3.5 and gives it medium to high capacity for performing sophisticated tasks, such as reasoning, translation and associations.
Latam-GPT is trained in a regional database, which compiles information from 20 Latin and Spain countries, with a total of 2645,500 documents impressive. The distribution of data shows a significant concentration in the largest countries in the region, and Brazil is a leader of 685,000 documents, and then Mexico from 385,000 Spain from 325,000, Colombia with 220,000 and Argentina with 210,000 documents. The numbers reflect the size of these markets, their digital development and the availability of structural content.
“Initially, we will launch a language model. We expect its performance in general tasks will be similar to large commercial models, but with excellent performance in the topics specific to Latin America. The point is that if we ask about topics important for our region, his knowledge will be much deeper,” explains Soto.
The first model is the starting point for developing a family of more advanced technologies in the future, including those with image and video, and for scaling to larger models. “Because it is an open project, we want other institutions to be able to use it. The group in Colombia could adapt it to the school education system or in Brazil can adapt it to the health sector. The idea is to open the door to various organizations to generate specific models for individual areas, such as agriculture, culture and others,” explains the director of Cenii.