This is the ultimate vision that our GER team wants to achieve, basic intelligence. Stephanie Zhan When it comes to virtual worlds and games, you've unlocked some of the thinking and emergent behaviors in open environments. What's your personal dream in the world of games? What kind of innovation do you hope AI agents will bring to the field of games? Jim Fan Yes, I'm particularly excited about two aspects. The first is the agent in the game. Today's NPCs (non-player characters) have fixed scripts and are all hand-written. But what if we had truly "living" NPCs? They are able to remember things you've said to them before and can take actions in the game world that change the narrative and the development of the story. This is something we haven't seen yet, but I feel like there's a lot of potential here.
different experience playing the same game, and even if a estonia phone numbers person plays the same game twice, the story will be different, so each game has infinite replay value. The second aspect is creating the game world itself. We've seen a number of tools that do some of this, like generating D assets from text, or models for video from text, and language models that can generate stories. If we combine these tools, the game world can be generated on the fly and interact with you as you play. It's going to be a really amazing and really open experience. Stephanie Zhan Extremely interesting. In terms of the vision of intelligent agents, do you think GPT-level capabilities are needed? Or can a model like Llama 8B achieve that? Jim Fan I think an agent needs to have a few key capabilities.
First, it needs to be able to hold interesting conversations, have a consistent personality, possess a long-term memory, and be able to take action in the world. In this regard, the current Llama model is pretty good, but not good enough to produce very diverse and engaging behavior.So I think there's still a gap in that regard. The second issue is the cost of reasoning. If we want to deploy these agents to players, we need either very low-cost cloud hosting or the ability to run on local devices, otherwise they're not cost-effectively scalable. So that's another factor that needs to be optimized. Sonia Huang Do you think that all work in the virtual world is to serve the goals of the physical world? Or is the virtual world a valuable enough field in itself? How do you balance your work priorities between the physical and virtual worlds? Jim Fan I think that the virtual world and the physical world are ultimately just different realities on the same axis.