Thanks for organizing this! I had a great time playing with the system for a week even though I did not get anywhere close to the prize pool.
I have no idea if anyone checks these threads after a week but not that it's over, I am super curious what strategies other folks used. I used a python script to pre-process my programs to minimize jumps and downloaded the sim to run an iterative optimizer with parallelization. I just put a little writeup and the code on github https://github.com/Will-Morr/MomentSwarm.
I just came back to check if anyone else was discussing their entry, so thanks for posting your write up!
I wrote mine by hand just testing it using the in-browser simulator. I initially used subroutines to try to keep things structured but then resorted to inlining most things (and flattening as much state as I could to bare registers) to try to speed things up, though i never quite got my ants to always move within the 64 op step.
My best score on the official test set was 654 (43rd place). I put my code and a little write up here: https://github.com/cristoper/brain.ant
Here is my entry, I managed to get up to 815 on the validation set during the event, and 825 after:
https://github.com/hukdukmukduk/brain.ant
Thanks for sharing, and congratulations!
Interesting that you didn't use dead reckoning at all to return to the nest. I didn't think I could get away with that. And somehow using a second channel to extend the trails never occurred to me.
Thanks to everyone in this chain for sharing! I've learned so much from your programs and was inspired to make my repo public and do a write up as well. I was 52nd place with a score of 599.
https://github.com/Taxes/ants
Wow. thank you so much for the writeup.
I'm just now (last few months) dabbling with agentic coding, but being an oldschool tech guy I had to dive straight into assembly. It was just too hard for me to level up my agentic skills given the deadline.
I wrote a... slightly substantial writeup: https://hallofdreams.org/posts/agents-and-ants/
Thanks!
This challenged coincided with my first dabblings into agentic coding. I wrote all my attempts in hand coded assembly, without doing any basic research. but your site is very entertaining and is reminding me of why the connection of ants/swarms cellular automata to gradient discovery to AI always blows my mind