Open-source Aya can produce, translate and classify text. It works in 101 languages. Cohere for AI—a non-profit lab backed by Toronto-based Cohere—built the large language model (LLM) with contributions from more than 3,000 researchers. (The Logic)

Talking point: Aya’s goal is to broaden access to generative AI’s breakthroughs “beyond first-class citizen languages,” wrote the 17 named authors of the paper that unveiled it. The most popular and widely used LLMs like those underpinning ChatGPT are mostly trained on English datasets, and studies suggest they tend to perform best when they can “speak” that inherited tongue. Regional or national AI developers like Paris-headquartered Mistral AI have launched models that are conversant in high-volume local languages like French, Italian and German. But that still leaves out people working in less-resourced languages from any potential generative AI benefits. In addition to Aya, Cohere for AI released its underlying dataset to prompt more R&D on that front.