In this age of technology, AI models are being developed every day, and some have even come to believe that AI is getting smarter. While the jury is still undecided on the merits of this claim, Meta has announced another language model that it believes is significantly better than the competition in translations.

The new AI model is called the NLLB-200; where “NLLB” stands for “No Language Behind” and “200” stands for the number of languages ​​it can translate. Meta has also open-sourced the model on GitHub and has a variety of options starting at 600 million parameters and going all the way up to 54.5 billion parameters depending on your use case.

Meta boasted that the NLLB-200 outperformed other AI studies in the field by an average of 44%. It was even 70% better for some Indian and African languages. These results are based on the FLORES-200 dataset created by Meta to be evaluated in 40,000 different directions of speech. The dataset is publicly available to developers.

Additionally, the company announced: We also provide grants of up to $200,000 to researchers and non-profit organizations with initiatives in sustainability, food security, gender-based violence, education and other fields that support the UN Sustainable Development Goals to benefit from the NLLB-200. Nonprofits interested in using the NLLB-200 to translate two or more African languages, as well as researchers working in the fields of linguistics, machine translation, and language technology are invited to participate.

Meta says this model could lead to improvements in many use cases, such as creating digital assistants or mechanisms for adding subtitles to movies. For its part, it plans to use the NLLB-200 on social media platforms (Facebook, Instagram) and metaverse development. Source