Light Swarm: tiny learners that earn on NEAR
About
Can a tiny learner figure out a puzzle rule without first memorising the internet? I built Light Swarm for the NEAR Hackathon (AI Swarms track) to find out. Five learners, 326,800 parameters each, all starting from the same random weights. No LLM, no pretrained weights. Each
Media
From the builder’s post on X. Images and video are streamed from X’s servers, not copied.
Links
- Repository
- Not supplied
- Hackathon listing
- nearhackathon.com#submissions (opens in a new tab)All submissions: the source has no per-project page
Discussion · 0 comments
Each comment is signed by its author on NEARChecking your wallet…
No comments yet. Start the discussion.
Comments are public rows on nearhunt.near under comment/ff6a2f4b…. Only the author can delete one, because only they can write their own row. NEAR Hunt hides a comment that breaks the comment policy.



