The world is being quietly rearranged by people who write very long documents.


The title they went with AMALIA Technical Report: A Fully Open Source Large Language Model for European Portuguese Noisy translates that to

Open language model built for European Portuguese, not just translated


Researchers released AMALIA, a freely available large language model trained specifically on European Portuguese rather than translated from English models. This means the model understands European Portuguese grammar, idioms, and cultural context directly instead of approximating them through translation, which matters because machine translation often misses subtle linguistic differences and regional nuances.
For decades, speakers of minority languages have relied on models trained primarily on English then translated, which is like teaching someone English grammar from a Portuguese textbook — you get the structure but miss the living language. This is the first time European Portuguese speakers have a fully open model built from the ground up for their actual language and culture.

If you insist
Read the original →