Aya - AI for all languages.

Cohere For AI just dropped a major multilingual AI bomb. LLMs are hot right now. But they often leave out a huge chunk of the world due to language barriers. Gotta think beyond English. Well, Cohere just did with Aya.

What's going on here?

Aya is a multilingual LLM that supports a whopping 101 languages.

What does this mean?

Most AI models get stuck on English and a few others, leaving a huge chunk of the globe out of the loop. Cohere for AI partnered with 3,000 researchers across the globe to create Aya.

Aya was trained on the largest multilingual dataset (513M prompts!) and outperforms other open-source multilingual models in tricky tasks like understanding, summarizing, and translating various languages. It also covers 50+ previously underserved languages; hello Somali and Uzbek!

The project is open-sourcing both the Aya model, as well as the multilingual instruction fine-tuned dataset.

Why should I care?

More multilingual AI powers up nuanced conversations across cultures, and build tools that work for everyone, not just some of us. Especially when most of us are working with global teams now. Okay, forget teams. You can finally explain your tech job to family (no promises though).

And, the Aya dataset is a treasure for multilingual research. You can contribute to the Aya project or use it to do some stuff on your own.