NVIDIA has announced a breakthrough in artificial intelligence for gaming at Gamescom 2024, introducing its first on-device small language model (SLM) designed to enhance digital human interactions in video games.
The new model, called Nemotron-4 4B Instruct, is part of NVIDIA's ACE (Avatar Cloud Engine) suite of technologies. It aims to bring digital humans to life with generative AI, providing quicker and more relevant responses for non-player characters (NPCs) in games.
"This new, optimised SLM is purpose-built with instruction tuning, a technique for fine-tuning models on instructional prompts to better perform specific tasks," said Ike Nnoli, an NVIDIA spokesperson.
The model is optimised for low memory usage, allowing it to run locally on over 100 million GeForce RTX-powered PCs and laptops. This local processing capability results in faster response times and more intuitive interactions between players and digital characters.
Mecha BREAK, a new game from Amazing Seasun Games, is the first to showcase this technology. In the game, players can converse with a mechanic character and instruct it to switch and customise mechs, demonstrating the model's improved role-play and function-calling capabilities.
NVIDIA's ACE technology combines several AI models for speech-to-text, language processing, text-to-speech, and facial animation. The modularity of ACE allows developers to choose specific components needed for their particular processes, whether running in the cloud or locally on RTX AI PCs and workstations.
As the gaming industry continues to evolve, NVIDIA's advancements in AI-driven character interactions promise to create more immersive and responsive gaming experiences. The company envisions applications for this technology beyond gaming, potentially transforming industries such as customer service, healthcare, and retail.