Latam-GPT: un modelo de lenguaje desde América Latina

·24 min·2 clips

LATAM-GPT is an AI model built from Latin American data to understand the region's language and context.

1. Lo Que Queda del Día on Radio Cooperativa covers LATAM-GPT's launch, Chile's productivity figures, and a bonobo symbolic-play study from Johns Hopkins. 2. Álvaro Soto, director of SEINA, explains the project and Pablo García, president of Chile's National Commission for Evaluation and Productivity, discusses the country's economic figures. 3. The episode's core question is what these two technology and economic developments mean for Latin America's ability to manage its own future rather than depend on decisions made elsewhere. 4. LATAM-GPT is described as the first large language model developed in Latin America, built over approximately two years with data from more than 65 organizations in more than 15 countries. 5. Unlike ChatGPT, which scraped internet data with unclear copyright attribution, LATAM-GPT was trained exclusively on data provided with explicit institutional consent — including full session records from the Colombian Congress. 6. Soto says the model is available as a chatbot but its primary value is as an open platform for entrepreneurs and academics to build region-specific AI applications in education, health, agriculture, and industrial centers. 7. An example application discussed is an educational tool that, instead of solving problems for students, would guide them through learning — functioning as a teaching assistant rather than an answer machine. 8. Soto notes large commercial AI companies will face additional licensing restrictions on LATAM-GPT data because they do not reciprocally share their own data. 9. Pablo García says Chile's total factor productivity grew for the second consecutive year in 2025, ending a decline that began around 2010–2012. 10. García attributes the productivity improvement to three factors: reduction in constitutional and pension-fund uncertainty after years of political instability; adoption of better technology and production instruments by Chilean companies; and successful diversification of export demand toward Asia. 11. Chile's construction sector is identified as the laggard, damaged by the withdrawal of approximately 60 billion dollars from pension funds, which reduced the savings pool available to finance mortgage lending and investment. 12. García frames Chile's current 2–2.5% growth rate against the 3.5–4% rates that historically successful economies achieved at comparable development levels, positioning Chile at the tail of OECD rankings. 13. The automation and AI discussion centers on whether technological change complements or substitutes Chilean workers, with García arguing upskilling and workforce adaptation — including reform of the SENCE labor training system — are the key challenge. 14. The bonobo Kansi segment describes a Johns Hopkins study published in Science in which a 43-year-old captive bonobo was put through tea-party experiments to test symbolic play. 15. Kansi correctly identified the location of imaginary juice and distinguished real grapes from plastic ones at rates statistically well above chance, according to the study's authors. 16. The narrator notes prior anecdotal evidence of chimpanzees playing with invisible objects or using sticks as dolls, but calls the Kansi result the most solid experimental evidence yet that symbolic imagination may not be exclusively human. 17. Kansi died in March 2025 at age 43, after the experiments were completed. 18. The segment maintains a reflective tone, noting that the capacity of animals to model the world mentally may be more complex than previously believed. 19. Curious listeners interested in Latin American AI sovereignty, Chilean economics, and animal cognition research will find each segment independently accessible. 20. Listeners expecting a single-topic deep dive or detailed technical AI coverage will find the episode's three-segment structure too light on each subject.

Listen to the show on

Apple Podcasts Spotify

Play clips from this episode More from Lo Que Queda del Día