Nvidia’s New Software Hopes to Stop AI Chatbots From Hallucinating

Nvidia has developed open-source software to help developers create AI applications that stay on message, solving the problem of hallucinations (which is when an AI just makes stuff up.) 

The appropriately-named NeMo Guardrails software is specialized to interact with large language models (LLMs) like ChatGPT and other popular AI chatbots. It works to make sure the responses from a chatbot are “accurate, appropriate, on topic, and secure.” The software suite includes all of the code that companies need to integrate chatbots into platforms safely and securely. 

What does this mean for you? There are a whole lot of AI chatbots coming soon to various apps on your phone, tablet, and beyond. Gigantic platforms like Snap and Bing have already implemented chatbots, but Nvidia’s software will allow smaller businesses to implement AI.

This could go beyond apps and into, for example, a bank’s phone system. Nvidia also says that they see this software being useful in a variety of industries, from designing new drugs for pharmaceutical brands to writing new software from scratch. 

NeMo Guardrails is open-source software, so enterprising users can easily make changes to suit their particular needs. It works by keeping an AI chatbot from veering into undesired areas and by ensuring it gives accurate and appropriate information. On a macro level, an undesired topic could be something controversial or icky, but on a micro level, it could just refer to subjects not in line with a particular company’s vision. In other words, an AI chatbot affiliated with a bank could talk to you about your money but not about that latest episode of Succession. 

There are also other guardrails in place to keep AIs from making connections to risky websites, favoring credible sources. Nvidia’s NeMo Guardrails software is available now, but it will be a while before we start seeing results within apps and services.