NVIDIA Unveils New Chatbot for RTX PCs

Nvidia’s ChatRTX Trains AI on Your PC for Local Search.

Nvidia is re-entering the “AI on your PC” market with ChatRTX, an enhanced chatbot application. In February, the organization introduced a technological demonstration known as Chat with RTX, which has since been modified and abbreviated. The new version is available for free download by anyone with an RTX GPU and includes enhanced functionality and new features. 

The primary functionality of ChatRTX is the capacity to train a substantial language model using local content stored on your personal computer. According to Nvidia, you could point it at a folder containing files and then ask queries about that content. Although this enables it to execute more rapidly than cloud-based applications such as Microsoft Copilot or ChatGPT, it cannot rely on petabytes of data to generate its responses. Additionally, by implementing the application locally, any data you input into it remains confidential, unlike when it is uploaded to the cloud. According to Nvidia, ChatRTX can ingest multiple file formats, including txt, pdf, doc/docx, jpg, png, gif, and XML. Therefore, you can direct it to folders containing those files and then query it regarding the content it finds. 

The application also supports speech-to-text and multiple languages, allowing you to pose queries to the bot via your computer’s microphone. Additionally, voice or text can be used to navigate through image folders; for instance, “hiking in the mountains” will retrieve images of you trekking. It is comparable to Windows search, which may operate effectively but with the added capability of identifying the content of the photos rather than relying solely on their file names. TechPowerUp reports that OpenAI developed the CLIP (contrastive language-image pre-training) technology, which enables it to accomplish this.

ChatRTX incorporates several newly added languages and updated models from the previous release. The Mistral 7B and Llama 2 models have been updated, and Google’s Gemma model and ChatGLM3 have been integrated; ChatGLM3 supports both Chinese and English. Its ability to select from various language models when executing queries is a significant feature that differentiates it from other chatbots.

Despite these advancements, it is reportedly designed to respond to single-topic queries but struggles with follow-up questions to previous inquiries. For instance, if you requested hiking photographs, it is not possible to enhance that search by including supplementary inquiries concerning the results it returned. This is due to the evident challenge it faces in comprehending the connection between questions and responses at this nascent stage. 

Despite this, the app is available for free download and is a staggering 11.6GB. It can be immediately downloaded from Nvidia’s website.