A dozen Latin American countries joined efforts to create the first large artificial intelligence language that understands the peculiarities of the region and whose first version will be launched in September, Chilean authorities announced on Tuesday.
The open source LATAM-GPT initiative is coordinated by the State National Artificial Intelligence Center (CENIA) of Chile in conjunction with more than 30 institutions throughout the region as collaborative space.
“We believe it can be a democratizing element of artificial intelligence and an accelerator of the use of AI in different spaces,” the Chilean minister of Sciences, Aisén Etcheverry, told journalists.
“In the case of Chile we are starting with the municipalities, but why not dream of schools, hospitals? (…) that have access to a language model that is not imported and reflects our language?” He added.
The project started in January 2023 with reduced financing of each participant in order to correct the usual inaccuracies and performance limits of other models generated in English and for the global context, those responsible detailed.
Lee: XAI of Musk negotiates capital financing of 4,300 million pesos, reveals Bloomberg
The authorities clarified that the model is a data engine on which applications and other solutions can be developed, such as well -known chatbots.
“LATAM-GPT is not comparable with Depseek or Chatgpt and the result of this is also very different because it is not a product that I am going to buy but is an engine that will be available,” said Etcheverry.
The Latin American model also seeks the conservation of native languages, since even the first translator of the native language of the Easter Island was generated, which they intend to extend to other ancestral cultures of the rest of the countries.
The intention with language is to develop virtual assistants for public services, personalized educational systems or translators of native languages. Pilots have already started in Chilean municipalities for local solution applications.
Technically, the model is a flame version 3 of 70,000 million parameters whose training is carried out in a regional supercompute structure including the center of the Chilean University of Tarapacá and cloud computing. It also had the support of the CAF Regional Development Bank.
Despite not having a budget for its impulse, the head of the Cenia, Álvaro Soto, is confident that as the system capacities are shown, more resources are accessed.
“I don’t see myself competing with what Chatgpt is doing or these companies that are putting amounts of money that really, however we get together, is not the amount of money we can raise,” he said.
Lee: These are the latest advances by Canvas AI, the ‘chatgpt’ for companies
The project also has an alliance with Amazon Web Services and the Data Observatory that enables it a greater capacity for data computing.
Meanwhile, CENIA is actively looking for unpublished data collection throughout the region to feed the model, although it acknowledges that the generation of data from some areas such as the Caribbean would take more time due to the digital divide that persists in many nations of the continent.
With Reuters information
Do you like to inform yourself for Google News? Follow our showcase to have the best stories
