Latin America has now been a cradle of inspiration behind staple foods and famous Happy Miles, globally popularly popular and music styles, potatoes. It can also become a new form of AI.
A coalition of research institutes is working on what they say to Latmagipt – a tool that can take into account the language version of the field, cultural experiences, and “idiocyticis”.
It aims to make users more loyal to the US and Caribbean’s larger language models (LLMs) that mostly come from the US or Chinese companies and were largely trained in English.
Rodrigo Duran Rojas, director of Chile’s National Center for Artificial Intelligence, said, “We want to develop our abilities, find local AI-based solutions and make a better understanding of these devices in Latin America and in Latin America and in Latin America,” Rodrigo Duran Rosas, Director of Chile’s National Center for Artificial Intelligence, said that the attempt is said.
Durán Rojas stated that for general purposes, the project would be barely suppressed to compete with the “status of the art model with the multimilian budget”, but “What can offer our models can offer that others are a very rich and representative approaches of Latin America and Caribbean,” its people and its output.
For example, Duran Rosas said that the initial testing has shown better results to the Latmagipt on knowing about South American history, and that LLM is asked to write a poem in the style of local writers or provide an observation of regional education policy.
The hemispheres include more than 30 institutions in developing latamgpt from countries, and colleagues include Latino in the US such as Freddy Villas Menceses, an associate professor of Hispanic studies at Lewis & Clarke College in Oregon. This, he said, “Latin and Latin American experiences are a cultural fellowship, which is beyond geography.”
“In Oregon, in California, in Texas there are elements of Latin America,” said Wilchens Menceses. “We also want to include that Latin experience.”
Latamgpt, which is aimed at launching its first publicly available version around June, was declared to focus on “moral, inclusive and beneficial” on the heels of a regional commitment made during a summit on artificial intelligence in Uruguay last month, which was declared to focus on “ethnic, inclusive and beneficial” technological development which will “promote and preserve human rights and protect” To detect policies.
It follows a growing lift in the field of impulse technological progress such as the use of drones to monitor the harvesting of forests in Amazon Rainews, the development of apps to encourage more people to learn indigenous languages, which creates algorithmas, which creates algorithmas, which creates algorithmas, which help people to be creating for the disappearance, which is a blockching system to stop historic documents to the disappearance. Helps adopt.
Some of those protected documents are now being used as sources to train Latamgpt, as well as institutions such as the papers, records and logs, as well as the library and National Archives, are specially made available to the project. Durán Rojas stated that it gives more nuances and localized widths to the model, which is to use other systems to use general internet data scraping.
“Latmgipat will have more reference to other model languages and therefore it should be very low” when it comes to matters of its use, Duran Rosas said. The hallucinations are the one who calls an AI researcher when a model appears that creates an answer that is wrong or wrong, although it is presented as factual.
So far the dataset of the project has more than 8 terabytes, so the model can run at about 55 billion parameters (with the variable that makes an LLM a prediction output, such as neurons that sink or connect in the human brain). Durán Rojas stated that the first public version of the slapping is somewhat close when Openai launched it in the fall of 2022.
Miscellaneous dialects and complex grammar challenges
Chhatgpt and other models like Google’s Gemini have also sought to include a comprehensive scope of data to offer programs in other languages in recent years – such as knowing the LLM to respond in the metric system when the relevant or idioms are understood.
Those companies acknowledged the importance of expansion of that offer. Hugong Cho, director of Engineering and Internationalization for Google’s Gemini apps, said it is a “dedicated experience” that may be necessary for cultural relevance and sensitivity. ”
But they also believe that this is a particularly complex effort, as most of the training data available for them is in English. Cho said, “The intricacies of different languages can create a significant obstacle for all AI models. … Languages with complex grammar, diverse dialects or limited digital resources can be difficult to train,” Cho said.
Latamgpt, through its institutional network with libraries and archives, skirts to some extent – but not completely. Durán Rojas said they are still struggling to include indigenous languages spoken by millions of people in the region because written documentation is not widely available.
But they are still trying because they constantly correct their models – although they emphasize the importance of cooperation.
Professor Viles Menses of Lewis and Clarke said, “The quality and characteristics of the results we can achieve depends on us as Latin Americans.
Currently, with the date of temporary June launch, latamgpt is still receiving data because allies regularly check with specific questions to be benchmark compared to other available models.
Many of the questions they are testing, many different names and words used in the field are a request for a specific term such as “car”, or GPT that a request for GPT to respond to large -scale immigration from places like Venezuela.
According to Duran Rosas, a major goal of laids is to be familiar with these technological progress to include them in public policies and rules.
For that, it is important to build a transcontinental network to help develop the project, and it will probably remain the same according to Duran Rosas.
“The most meaningful aspect, the greatest heritage, it is a mutual relationship that we have found to strengthen and develop AI-based solutions,” they say. “Model, I mean it’s great that we are making it, but cooperation – that’s the most impact will be how we create things.”
And at the same time there is an increasing opportunity to contribute further with a Latin touch.
“On its base, it is jointly making something from Latin America to Latin America and the world, for ourselves and for others, which we can also produce high -tech,” said Vilesus Mences, “And we can contribute to the knowledge of artificial intelligence while employing our social and cultural intelligence.”
A pre -version of this story was first published by the Notice of Telemundo.
