Latin America is embarking on a bold technological adventure with LatamGPT, an artificial intelligence model designed to reflect the region’s identity, languages, and cultures. Scheduled for launch by the end of the year from Chile, this project aims to provide Latin America with a unique AI capable of understanding and representing its specificities, moving away from the dominant models designed in the United States or Europe.
LatamGPT: An AI rooted in Latin American reality
Unlike large models like GPT-4, trained mostly on data from English-speaking or Western contexts, LatamGPT aims to capture the essence of Latin America. This project, led by the National Center for Artificial Intelligence (Cenia) and the Data Observatory (DO) in Chile, focuses on collecting and processing data in Spanish, Portuguese, and English from various sources such as government institutions, universities, and digital platforms.
The goal: to create a model that integrates cultural richness, indigenous knowledge, and local specificities, which are often absent or poorly represented in current AIs.
A technological and cultural project
LatamGPT is not just a technical feat. According to project leaders, it is a tool to assert technological sovereignty and promote a vision unique to Latin America. Foreign models, while effective, struggle to grasp the nuances of regional realities, which can lead to biases or inappropriate responses.
With LatamGPT, the idea is to give a digital voice to the region, capable of understanding its languages, histories, and social contexts.
Read also on this topic: Fouju: a French village transforms into a global AI hub
LatamGPT: A titanic undertaking
The development of LatamGPT relies on a colossal effort of data collection and processing. Currently, the project is in its third phase, which involves sorting and classifying quality information from public sources, such as academic articles, blogs, and educational resources.
About 500 GB of data in Spanish and Portuguese has already been collected, with a final goal of 20.5 TB, including English data from databases like RedPajama v2. This corpus, rich with billions of documents, covers fields as varied as economics, social sciences, arts, and medicine.
The project benefits from the support of Amazon Web Services (AWS) for the computing power required for this undertaking.
The Data Observatory plays a key role by processing massive volumes of data, while Cenia ensures its quality and relevance. This collaboration aims to produce a robust, reliable, and representative model.
Toward a regional technological revolution
LatamGPT embodies a strong ambition: to position Latin America as a major player in the global AI landscape. By developing a model that reflects its realities, the region aspires not only to fill the gaps in current AIs but also to inspire other territories to claim their place in the technological revolution. With this project, Latin America is not just following: it wants to lead the way.
Source: Latercera
