I plugged a local 2B decision model into Halo CE for split-screen co-op ๐
retained text
So i got bored and asked GPT to plug Intern-Decision-2B into OpenCE. It gets the game state as text, player/NPC positions, health, ammo, etc and some raycast to help it detect obstacles and it then predicts actions that are translated into game input. I also added an in-game text box so I can prompt it while playing. It has a short history of its recent actions and an aim-assist toggle. I tried multiple iterations, getting a few decisions per second depending on the setup, running locally on my RTX 4060 laptop. I love how the whole experience somehow works while also being completely broken. I had a lot of fun messing around with it and trying some abstract instructions as you can see ๐
document fetched 2026-10-14 00:30 ยท canonical https://reddit.com/r/singularity/comments/1x5bcby/i_plugged_a_local_2b_decision_model_into_halo_ce