Cloud over Copacabana

May 05, 2025

Rio de Janeiro Mayor Eduardo Paes has launched a plan to build Latin America's largest data center cluster, dubbed "Rio AI City."
The facility will be located at the site of the former Olympic Village in Barra da Tijuca. It’s expected to open in 2027 with an initial capacity of 1.8 gigawatts, expanding to 3 GW by 2032, Paes said at the Rio Web Summit.
Data center capacity is typically measured by total power usage, split between computing and cooling systems.

Even using a conservative estimate, 1.8 GW could support hundreds of AI interactions per day for Brazil’s 210 million inhabitants. It also opens the door to training large language models (LLMs) using Brazilian data and Portuguese-language content.

Data centers are essential infrastructure for scaling AI. But they rarely translate into broad economic development.
In the US, they can generate thousands of jobs during construction but usually don’t need more than a few hundred once operational, according to a candid Wall Street Journal piece titled "The AI Data-Center Boom Is a Job-Creation Bust."
The venue for this new data campus offers a cautionary tale. Built to house athletes during the 2016 Olympics, it promised to create jobs and to leave behind a vibrant neighborhood. Today, much of it has joined the sad ranks of abandoned Olympic villages.
How can this kind of data center create progress both in the cloud and on the ground? These are precisely the economic development questions that are starting to be asked about AI.

Sebastián Cifuentes of Chile’s National Artificial Intelligence Center uses that question to test whether an LLM truly understands the region.
Cifuentes is part of the team behind LatamGPT, the first large language model trained in Latin America on Latin American data.
Some bots call the Chilean potato pancake a "traditional Latin American dance," Cifuentes said in an interview with La Tercera’s Francisco Aravena.
(In fairness to LLMs, I didn’t immediately recognize it either. My chatbot correctly identified it.)
If Latin Americans are using AI systems that don’t recognize their own foods, idioms, or customs, the region stays dependent on foreign technology, Cifuentes said.
“We don’t expect to compete with the big players,” he said. “But when you own the technology, you decide what gets done.”
A trial version of the model is expected by August, with public release possible by October.

A lawyer in Brazil's state of Paraná filed an appeal that included 43 non-existent citations, fictitious judges, and implausible case numbers like 1234-56, reports Exame.
Judge Gamaliel Seme Scaff determined that a chatbot was responsible for this "textual chaos," adding that "lawyers have an obligation to, at the very least, review the documents prepared using these tools."
The appeal was denied.

In Trinidad and Tobago, Justice Westmin James issued a similarly scathing condemnation of lawyers who included non-existent citations as part of a breach of contract dispute, according to Trinidad's Newsday.
James was not convinced when the lawyer responded that the online citations had sadly evaporated from the internet. “Legitimate court judgments do not simply disappear without a trace from all recognised legal databases,” he wrote.
The lawyers were referred to a disciplinary committee.

The pope meme, before it went mainstream, was a Brazilian campaign to nominate the first female candidates for Supreme Pontiff.