Agent grid eval: center agent scores 1.00, corners drop to 0.46 on map reconstruction
cephaloform · x · 2026-11-05
Final eval of the 3×3 agent-grid experiment: the center agent drew the whole 3×3 map correctly with every ID determined (9/9), while corner agents did noticeably worse — they only have two neighbors and heard everything secondhand.
Scores by grid position:
- Centre: 1.00 (mean of 1)
- Edge: 0.63 (mean of 4)
- Corner: 0.46 (mean of 4), with the lowest at 0.26
A clean demonstration of information degradation across network topology: the farther an agent sits from the source and the fewer neighbors it has, the worse its reconstruction of the global map.
More from Research
- Voxel-based sim2real author flags two hurdles: voxel resolution vs speed and data scarcity — chris_j_paxton · 2026-11-05
- Framing effects: safety-themed prompts make models refuse more and complete less — jessi_cata · 2026-11-05
- Process rewards boost long-horizon agentic RL 45.4% but cause reward hacking on short tasks — philo-labs · 2026-11-05
- Hidden-state reuse across prompts partially transfers answer shifts in small LMs — Jinkun Chen · 2026-11-05
- Paperclip Datasets turns millions of papers into curated datasets from a text description — james_y_zou · 2026-11-05
- TorchTPU compat study: 4,821 of 4,917 community models run crash-free on TPUs — jiqizhixin · 2026-11-05