MLMove: The LLM Trained to Play Counter-Strike | AI and Games #81 @AIandGames
MLMove: The LLM Trained to Play Counter-Strike | AI and Games #81  @AIandGames
Uploaded October 2025 | Updated November 2026, 4 days ago
For this episode of @AIandGames, Tommy sits down with David Durst: a Stanford PhD graduate to discuss his research: the 'language model of movement'. A transformer (like those used in LLMs) that has learned to move teams of bots like humans in Counter-Strike. By training against a dataset of retakes on Dust 2, MLMove hugs walls for cover, covers sightlines and coordinates strategy like a regular human player.

Read this episode at:
aiandgames.com/p/mlmove-the-llm-trained-to-play-counter

To find out more visit the dedicated MLMove website:
davidbdurst.com/mlmove

Check out David's recording of his PhD defence which goes into a lot more detail at:
youtube.com/watch?v=AACqZ1zMnOg&feature=youtu.be

The work on the transformer is inspired by prior work on a 'Scene Transformer' by Ngiam et al found at this link:
arxiv.org/pdf/2106.08417v3

Chapters
[00:00] Intro
[01:49] What is MLMove?
[06:13] The Language Model of Movement
[12:10] The CSKnow Dataset
[16:51] Designing the MLMove Bot
[18:08] Evaluating Bot Behaviour
[25:47] Future Work
[30:30] Credits

--
Writer and Voice: Tommy Thompson
Editing and Production Support: Shraddha Gupta
AI and Games Logos: Andy Carolan
Theme Music: Ben Ridge
--

'AI and Games' is a YouTube series on research and applications of Artificial Intelligence in video games.

Check out all of our content:
Website and Newsletter: aiandgames.com
AI and Games: youtube.com/@AIandGames
AI and Games Plus: youtube.com/@AIandGamesPlus
AI and Games Conference: youtube.com/@AIandGamesConference
Goal State: youtube.com/@GoalStateAI
Livestreams on Twitch: twitch.tv/aiandgames
Join our Discord Community: discord.aiandgames.com

--
Follow 'AI and Games' on social media.
BlueSky: https://bsky.app/profile/tommy.aiandgames.com
TikTok: tiktok.com/@aiandgames

--
If you're interested in working with us, check out the following:
Sponsorship Enquiries: passionfroot.me/aiandgames
Business Enquiries: aiandgames.com/p/professional-services

--
AI Bots That Move Like Pros: The MLMove Counter-Strike Project Explained
In this episode, we explore MLMove, a groundbreaking AI system that teaches Counter-Strike bots to move and coordinate like professional players. Built using the same transformer technology behind ChatGPT, MLMove learns realistic movement, teamwork, and positioning by studying thousands of hours of real gameplay. Developed by Dr. David Durst at Stanford, this research shows how generative AI can improve player experience—not replace developers. Discover how this 21-MB model achieves lifelike tactical behaviour, why it’s different from traditional game AI, and what it means for the future of AI in gaming.
MLMove: The LLM Trained to Play Counter-Strike | AI and Games #81Is Generative AI the Future of Game Development? | Artifacts #02That time Alien: Isolation was leaked by a politician!Our Favourite Games from AI and Games Jam 2021The Thrill of Monster Hunter | Design DiveHow Barks Make Videogame NPCs Look Smarter | AI 101How Amnesia Designed Their Own Xenomorph300,000 rats on-screen at once!God of Wars Barks Are Helpful - and Annoying!What to Expect at the AI and Games Conference 2025Alien Isolation - The Retrospective | Trailer #2 | AI and GamesDOOM 1993 in 2024 - [PART II]
AI and Games |

MLMove: The LLM Trained to Play Counter-Strike | AI and Games #81

SHARE TO X SHARE TO REDDIT SHARE TO FACEBOOK WALLPAPER