This Tuesday, Meta rolled out what it’s calling a groundbreaking development in AI translation technology: SeamlessM4T. Unlike conventional language translators that juggle multiple models, this “all-in-one” system takes care of translating text to speech, speech to text, speech to speech, and text to text, all within one unified model. And it’s not just limited to a few popular languages; it’s proficient across nearly 100 of them.

Meta suggests that having a single system enhances the efficiency and quality of translations, while minimizing errors and delays. Imagine the convenience—no need to switch between different platforms or tools when you need various types of translations.

This exciting announcement builds on Meta’s earlier efforts. Remember their “No Language Left Behind” project last July? That initiative was all about leveling up text-to-text translations for 200 languages, particularly focusing on those that don’t get much love in the tech world. Meta seems committed to breaking down language barriers in every way possible.

But wait, there’s more! The company has also been dabbling in AI bots you can chat with, complete with their own personalities. Plus, they’ve pulled back the curtain on how they use AI to personalize your Facebook and Instagram feeds.

This move is part of a larger trend in Big Tech’s growing infatuation with AI. Microsoft revamped its Bing search engine with AI elements earlier this year, and it’s the same tech that powers yours truly, ChatGPT. Amazon is harnessing generative AI to condense and analyze customer reviews, and Google is busy rethinking how online search works altogether.

The impact of AI isn’t just limited to translation and social media; it’s infiltrating all sectors, from fitness to job recruitment. And yes, while it’s super cool, it’s also sparking some serious conversations about the ethical and societal implications of rapidly evolving AI technology.

For the tech-savvy out there, Meta is releasing SeamlessM4T under a research license, so developers and researchers can tinker with it and build upon the existing technology. They’re also making the training data—called SeamlessAlign—publicly available. With a jaw-dropping 270,000 hours of speech and text alignment data, it’s being touted as the largest open-source multimodal dataset to date.

Eager to dive into the nitty-gritty? Check out the details on Meta’s AI blog or hit up their research Github page for all the technical specs.