Khaberni -Latin America has decided to carve its own path, away from global artificial intelligence models, notably ChatGPT.
The story began when Chilean graphic designer, Juan Palma, asked ChatGPT for simple directions to the nearest metro station from his home in Santiago.
However, the response was completely incorrect, leading him in the opposite direction.
This simple incident revealed a deeper flaw: global models do not properly understand local nuances, whether in language or culture, according to a report published by "restofworld" and reviewed by "Al Arabiya Business".
Latam-GPT: A distinctly Latin project
In response to this gap, a group of over 30 institutions in Latin America launched the "Latam-GPT" project, an open-source artificial intelligence model under development for two years, with a public launch scheduled for next September.
Hector Bravo, head of innovation at the Chilean company SONDA, says: "We are building artificial intelligence from Latin America for Latin America."
He adds that the project not only focuses on accuracy and response speed but also cares about cultural representation, social impact, and broadening access to artificial intelligence.
Forgotten languages and dialects come back to life
Latam-GPT is notable for its support of indigenous languages, such as Nahuatl, Quechua, and Mapudungun, in addition to rare local dialects from the Caribbean region.
This approach is similar to what is happening in Asia with the Sea-Lion project, in Africa with UlizaLlama, and in India with BharatGPT.
Although Latin American countries have been slow to adopt artificial intelligence, Chile has emerged as a leading force in this field.
Since the creation of the National Center for Artificial Intelligence in 2021, strategic alliances have been established involving over 50 billion "training parameters," equivalent to the capabilities of ChatGPT 3.5.
Local strength against tech giants
While models like GPT and Llama support Spanish, they often rely on translated texts or are tailored for Spanish culture, limiting their understanding of Latin contexts.
In contrast, Latam-GPT depends on local data from schools, universities, companies, and libraries, which enhances its capability for real interaction.
Data from "DemandSage" shows that Latin America, particularly Brazil, has become one of the largest users of artificial intelligence tools, enhancing the need for a model that considers regional specifics.
Technical, environmental, and legal challenges
The path is not without obstacles; the project requires substantial infrastructure and specialized expertise.
The use of energy and water raises concerns amid local opposition to data centers. The Latam-GPT team has stated that they use solar energy and flexible cloud infrastructure to minimize environmental impact.
On the legal side, data privacy laws vary from country to country, which could lead to legal complexities, especially with widespread use of personal data.
Representation of marginalized communities
The biggest challenge remains ensuring the involvement of indigenous, immigrant, and marginalized communities in the model's development.
The absence of these groups from the design of artificial intelligence could reproduce historical biases.
Officials believe that Latam-GPT is not just a technical project, but a demonstration of Latin America's capabilities to enter the world of artificial intelligence through its broadest doors.
