student projects on LLMs and games
Student Projects
Students on the Games Engineering MSc at Warwick University in the UK are kicking off their dissertation work this month. Im supporting a number this year. Its great work partly because you get to study several super interesting topics and also its nice to see the students rise to a new level.
LLMs and Game Agents
An area of interest for the current cohort are using LLMs in video game NPCs.
So here is a paper from Georgia Tech A Survey on Large Language Model-Based Game Agents which should be invaluable to any student looking into this area.
The authors survey the many recent papers and break down LLM-based game agents into three essential components:
- Perception: how the agent observes the game state
- Memory: mechanisms for short- and long-term information storage
- Reasoning & Action: the language model’s decision process and interaction interface with the game world
Overall it looks very like the Agent Based architectures from when AI was last popular back in the 80s and 90s and being done in LISP and Prolog. The main difference being that the LLM can do the reasoning and be the glue between the different parts of the agent. The Flexability of LLMs means they can handle a much more varied environment and react in more realistic ways.
A more specific and popular paper is Park et al 2024 Generative Agents: Interactive Simulacra of Human Behavior
I’d totally reccommend anyone who is interested in this exciting area of video game work to read both these papers.
( see also: msc game student llm npc ai )