Kyutai, a French startup, has launched an advanced artificial intelligence assistant called Moshi that sets a new standard for real-time AI conversations. This new assistant, which will be launched in July 2024, challenges the dominance of Silicon Valley by offering new opportunities for the development of communication with AI.
Founded in November 2023 with an investment of 300 million euros, Kyutai introduced Moshi to demonstrate Europe's competitiveness in AI development. The AI assistant, which is able to process and generate audio and text simultaneously, operates with a latency of just 200 milliseconds, which is significantly lower than other leading systems.
Technology
Moshi uses a language model called Helium, which consists of 7 billion parameters to enable real-time interactions. It was trained using 100,000 synthetic conversations and 20 hours of voice recordings of an actress. Moshi can express 70 different emotions and styles, providing a nuanced and human-like interaction experience.
The startup is supported by prominent personalities such as Xavier Niel, founder of Iliad, and Rodolphe Saadé, CEO of CMA CGM. Eric Schmidt, former CEO of Google, and Patrick Pérez, CEO of Kyutai, bring a lot of experience and credibility to the project.
Kyutai is characterized by its commitment to open source development. The company plans to release the code for the language model, the audio codec and the entire technology stack, inviting collaboration, like Sentient Labs, and global innovation. This transparency addresses ethical concerns and promotes community-driven progress.
Moshi's appearance underscores Europe's role in shaping the future of technology. The startup plans to further develop Moshi based on user feedback and interactions. Versions 1.1, 1.2 and 2.0 are already in development.
Moshi is designed to work on standard devices such as laptops and cell phones without relying on external servers. This device-based processing emphasizes privacy and security, as sensitive data does not need to be transmitted over the Internet. This makes Moshi very accessible and practical for everyday use as it integrates seamlessly with different devices and platforms.
