Nvidia AI tools could add toxic NPCs to your PC games of the future

gamesradar.com:

Nvidia is all about AI at the moment, but its next project could help make NPCs in your favorite adventures more toxic. That’s thanks to a new custom toolkit that “aims to bring intelligence “to non-playable characters through AI-powered natural language interactions,” using an approachable interface. Simply put, developers will be able to instantly fill games with a bunch of walking red flags, and I’m slightly terrified at the prospect. 

Unveiled at Gamescom, the customizable NPC tools are part of Nvidia ACE, the graphics card giant’s custom AI language model. In a press release, the company explains that LLM (large language models) are designed to only provide “ideal” responses, and there’s a distinct lack of emotion behind interactions. This is something you’ve probably noticed if you’ve ever tried to have a conversation with a chatbot, and it’s ultimately the thing that makes AI sound like an algorithm-fueled machine.

To combat this, and potentially turn non playable characters into absolute monsters, the newest interaction of ACE adds something called NeMo SteerLM, which allows personality attributes to be “quickly configured through simple sliders.” In a demonstration, Nvidia hooks up ‘Jin the ramen shop owner’ to its AI model and plays around with parameters using a sidebar. While options cover a spectrum of emotive traits, including humor and creativity, it’s hard not to think about the toxicity slider and how it transforms Jin into a nasty piece of work. 

The above flowchart helps outline exactly how NeMo SteerLM works from start to finish, covering the journey of an AI generated response to your voice input. Effectively, the whole setup uses selected attributes to determine an specific response based on NPC