Nvidia has showed off its own generative AI capabilities at Computex Taiwan. This demonstration was built by Nvidia in partnership with Convai and is aimed at showing off the suite of tools allowing game developers to integrate generative AI into NPC dialogues, meaning instead of pickingd for a predetermined dialogue tree, you can now actually talk to in-game characters.
The suite is called Nvidia ACE (Avatar Cloud Engine) and can run both on the cloud and locally. The suite comprises of Nvidia’s NeMo tools for deploying large language models, Riva speech-to-text and text-to-speech and several other smaller tools that bring everything together.
As for the demo itself, it’s built in Unreal Engine 5 with a healthy dose of ray tracing. Honestly, the demo itself goes a long way to show just how much Nvidia has stepped up its game when it comes to rendering photorealistic graphics. However, while the visuals are stunning, the dialogue itself leaves much to be desired.
What Nvidia is calling a “peek at the future of games”, is just you talking to an NPC characters, instead of picking dialogues from an existing list. The replies we saw in the demo aren’t as good as what we’ve come to expect from LLMs like GPT-4 and PaLM 2. But the important part here is to remember that the NPC is reacting to natural language. If implemented properly, this does have massive potential in games.
Nvidia hasn’t launched the demo publicly at the time of writing, so it’s hard to say how the model will behave when faced with all sorts of random questions outside of a controlled environment. However, The Verge reports that it can be scaled up to more than one character at a time, theoretically allowing even NPCs to talk to each other. Although this hasn’t been tested yet.
The fact that the demo was made in Unreal Engine 5 seems to indicate that a plugin for the engine might soon be on its way. That said, it’s also not clear at the moment just how eager game developers are to integrate this tech into video games, where in case the model goes unhinged, it’ll be more complicated to fix than simply restricting its answering capabilities.
In the News:Â Arm announces new CPU and GPU designs alongside Cortex-X4
