Meta has developed an AI model called SeamlessM4T, which can translate and transcribe nearly 100 languages in both text and speech. According to Meta, this represents a significant breakthrough in AI-powered speech-to-speech and speech-to-text. The model eliminates the need for a separate language identification model and aims to improve communication among people who speak different languages. Other companies, including Google and Mozilla, are also investing in AI translation and transcription tools. SeamlessM4T combines translation and transcription capabilities in a single model and was trained using publicly available text and speech data.

However, concerns about biases in AI-powered translations and transcriptions have been raised. Meta acknowledges that the model overgeneralizes to masculine forms and can produce toxic translations in certain languages. It is also important to note that AI translations may lack the accuracy and richness of human interpretations. As a result, Meta advises against using SeamlessM4T for long-form or certified translations, as well as for medical or legal purposes. Nonetheless, Meta believes that its model can contribute to a more connected and understood world.