One morning in May, Juan Palma, a graphic designer in the Chilean capital of Santiago, asked ChatGPT for instructions on how to get from his home to a nearby subway station. Within seconds, the chatbot spat out a response — which sent Palma in the opposite direction.
ChatGPT, among the world’s most popular generative artificial-intelligence models, was out of its element in the South American nation, Palma realized. “I was surprised by the orientation error in ChatGPT’s response,” he told Rest of World. “I had different expectations for the accuracy of this tool.”
While large language models, including GPT and Meta’s Llama, are trained on a wide range of data in languages other than English, their capability in those languages remains limited, particularly in dialects and local idioms. To address these shortcomings — which have led to inaccuracies and hallucinations, or fabrications — a group of over 30 institutions across Latin America has spent the last two years developing Latam-GPT. The open-source LLM, which will be available to the public in September, is being trained by locals who take language and cultural nuances into account.
The Chile-led Latam-GPT project is “building AI in Latin America, for Latin Americans,” Héctor Bravo, lead of disruptive technologies at Sonda, a Chilean IT firm that is not involved in the project, told Rest of World. “It means redefining success metrics — not just accuracy or speed but cultural representation, social impact, and accessibility.”
Latam-GPT is being designed for deep multilingualism and includes Indigenous languages, such as Nahuatl, Quechua, and Mapudungun, as well as dialect variants, including some from the Caribbean, said Bravo.
The Latam-GPT project is building AI in Latin America, for Latin Americans.
Latin America is following the lead of other regions. Southeast Asia’s Sea-Lion is a family of open-source LLMs trained in nearly a dozen regional languages besides English. In Africa, users can interact with UlizaLlama in at least five different languages, including Xhosa and Zulu. While, in India, BharatGPT supports over 14 regional languages, with the government recently announcing that it was building its own LLM as well.
Latin America has been slow to adopt AI. It is beginning to catch up, however, with Chile leading in terms of regulation and institutional development, according to the Atlas of Artificial Intelligence for Latin America and the Caribbean, a 2025 report from the United Nations Development Programme. Chile’s National Center for Artificial Intelligence (CENIA) was founded in 2021. The idea for Latam-GPT emerged shortly after.
“Given the scale of the initiative, we knew it was something that required broad collaboration from many stakeholders,” Alvaro Soto, head of CENIA, told Rest of World. His team wanted to create something “with an open spirit” and began putting together databases to feed the project, recruiting universities, government offices, and civil organizations, he said.
Last year, CENIA signed 33 strategic alliances for Latam-GPT across Latin America, the U.S., and Spain, eventually assembling 50 billion parameters — the equivalent of ChatGPT 3.5.
Although LLMs such as GPT and Llama 2 support multilingual capabilities, including Spanish, many of the data sets they are trained on are from Spain or translated from text originally written in English, limiting their ability to understand cultural and linguistic nuances. Latam-GPT, which is being trained with data from schools, businesses, libraries, and historical texts, “helps the model better understand the contexts and needs of Latin American users,” Omar Florez, Latam-GPT’s technical lead for pre-training, told Rest of World.
There is increasing demand for generative AI platforms in the region. Brazil has the highest number of users of ChatGPT after the U.S. and India, according to DemandSage, a sales analytics platform, and Llama downloads have also surged in Latin America. Teachers and students use them in classrooms, while business owners turn to them to offer customer support. Even government offices employ them to reduce processing times. In Buenos Aires, for example, the courts use ChatGPT to draft legal decisions.
Clearly, the resources behind ChatGPT dwarf those of Latam-GPT, which will be text-only for the foreseeable future. It will also lag behind on general questions and those not related to Latin America, said Soto.
Latam-GPT “requires ultra-high-capacity infrastructure, specialized talent, and relevant data sets — three areas where gaps still exist in the region,” Carlos Honorato, chief executive officer of Orión, a Chilean AI company, told Rest of World.
Still, the project “represents a strategic start to narrow the AI gap” with the northern hemisphere, Carlos González, deputy director of the IT and telecommunication department at Duoc UC, a private institution of higher education in Chile, told Rest of World.
To be successful, Latam-GPT will need to ensure the participation of Indigenous peoples, migrant communities, and other historically marginalized groups.
There are other challenges, too. Across the world, environmental experts have warned about the long-term impacts of LLMs, which typically consume significant amounts of energy and water. In many countries — including Chile — locals have pushed back against data centers, which house the infrastructure required to train and build these AI models.
The computing infrastructure for Latam-GPT is housed at the University of Tarapacá, in northern Chile — a region that’s been pummeled by a drought for decades. But while “training such a model for [even] 40 days requires the equivalent of the electricity used by thousands of households,” it still represents only a miniscule fraction of the total energy consumption in the country, Danilo Naranjo, the chief executive officer of Wingsoft, a Chilean software development and cloud consulting company, told Rest of World.
The team at CENIA said they use a flexible and scalable cloud-based infrastructure that optimizes resources and reduces energy use. They also use solar energy, which should significantly limit its environmental impact.
Legal analysts also worry about the patchwork of data privacy legislation in Latin America, which could lead to litigation and sanctions. Brazil, for example, is home to robust legislation, while neighboring Bolivia lacks comprehensive personal data protection laws.
“These types of issues can result in significant negative reputational impacts stemming from the mishandling of personal information,” Ricardo Lillo, a professor at the Adolfo Ibáñez University, told Rest of World.
Despite its claims on representation, some experts worry if the homegrown LLM will represent minorities accurately — and how these groups will get access. While its design marks a step forward from global models, data availability is still a barrier, Varinka Farren, chief executive officer of Hub APTA, which promotes Chile as an innovation center, told Rest of World.
To be successful, Latam-GPT will need to ensure the participation of “Indigenous peoples, migrant communities, and other historically marginalized groups in the model’s validation,” said Farren.
This is one of the goals for Latam-GPT, Rodrigo Durán, CENIA’s general manager, told Rest of World. While initial testing has been encouraging, it will likely take at least a decade to meet this goal, he said.
For him, the greatest contribution of Latam-GPT “will be demonstrating that we — Latin America and the Caribbean — have the capabilities, the talent” to carry out such an ambitious project, Durán said.
Great Job Cristián Vera-Cruz & the Team @ Rest of World – Source link for sharing this story.
