Nvidia unveiled a prototype of its R2X AI assistant at CES 2025, which runs on PCs and runs directly on the machine’s desktop. The assistant looks like a character from a computer game and helps you navigate through applications. You can connect any of the popular large language models to the assistant, including OpenAI GPT-4o or xAI Grok.
Nvidia AI models are responsible for the visualization and animation of R2X. The user can communicate with R2X in a text chat or in voice format, it is possible to upload files to the application, and it is possible to broadcast an AI image from a computer screen or from a camera. With this project, Nvidia aims to combine generative AI technologies in games with advanced large language models – ideally creating an AI assistant that looks like a person.
Here’s Nvidia’s R2X, but powered by Grok pic.twitter.com/kyOOORQ1kR
—
The company intends to open source the project in the first half of 2025. Nvidia is positioning it as a new user interface for developers, allowing end users to connect their favorite AI products from the cloud or on-premises. Modeled after Microsoft’s Recall feature, the R2X app can also continuously take screenshots and analyze them using AI, but this feature is disabled by default. If you activate it, the system will help you understand the software on your computer or, for example, give advice when developing complex program code.
—
In practice, Nvidia R2X is not working perfectly yet. During the CES 2025 demo, the avatar would sometimes exhibit an “uncanny valley” effect, where the character’s face would freeze with a strange expression; and his tone sometimes gave the impression of being aggressive. The AI assistant’s advice was mostly useful, but there were some “hallucinations” – he got confused in the functions of Adobe Photoshop, and then suddenly stopped “seeing” the screen image. In another demo, he compiled a summary of the contents of an uploaded PDF file.
Here’s R2X helping us use generative fill in Adobe Photoshop (it gave us incorrect instructions though) pic.twitter.com/CDLjbduBEw
—
To animate facial expressions in a conversation, the Nvidia Audio2Face-3D AI model was used, which did not always work perfectly. In the future, R2X will be able to participate in Microsoft Teams group video calling sessions and even act as an AI agent, performing certain actions on the computer desktop.