National LLMs to Save Languages | Google Machine Learning Certification Free | | | Turtles AI

TAILORED AI : PATRIOTIC LANGUAGE PATTERNS ON THE RISE

Generative AI brings many negative effects, including linguistic flattening. Tools such as ChatGPT and Gemini often produce texts with recognizable patterns and translations that lose cultural nuances, coming across as impersonal and uninteresting.

To counter this phenomenon, several countries are developing their own linguistic AI models ( LLM ), adapted to national linguistic and cultural peculiarities.

Recently, "Italy," a model created by iGenius and Cineca, trained on a large corpus of Italian words, was launched. Although it has fewer parameters than GPT-3 and GPT-4, the importance lies in its inferential capacityand the quality of the training data.

"Italy" is available open source for research and business and can be used via web interface or API. Editoriale Nazionale contributed to its training by providing historical archives of articles.

In Europe, French startup Mistral AI has developed the "MIstral" model, which is distinguished by being open-weight, allowing anyone to modify it, and right now it is one of the most powerful in Europe.

Globally, China is very active with more than 40 LLMs approved for public use, while in the United Arab Emirates "Falcon," a model focused on the Arabic language and capable of processing text, images and audio, has been created.

Universities are also involved in the development of LLMs. For example, Fabrizio Silvestri of La Sapienza University in Rome developed "Dante," which is based on Google transformer models and uses recent innovations such as weight quantization. The model wasaddressed in Italian with the help of the "Mistral" model.

These developments show a trend toward the creation of national LLMs aimed at preserving the linguistic and cultural characteristics of each country.