IAAICoT Monitorability of LoopLMs
On the Chain-of-Thought Monitorability of Looped Language Models
The first systematic evaluation of whether looped language models remain chain-of-thought monitorable, varying loop depth within a family and comparing against non-looped models matched by size, layer count or effective depth.
Inside the method
[ Shared layers ] × R → visible CoT → monitor
Simplified conceptual schematic. Consult the paper for the complete architecture.
- Recurrence family
- Full-stack recurrence
- Depth control
- Varied loop depth
- KV / state strategy
- See paper
Reading note
Deeper loops reduce monitorability on some MonitorBench tasks under stress tests, not fully explained by difficulty or length. Cross-model comparisons find no evidence that looped models are systematically less monitorable.
Sources checked 2026-10-28. This catalog does not imply independent reproduction.
Related works
Cite this work
@misc{cotmonitorability2026,
title = {On the Chain-of-Thought Monitorability of Looped Language Models},
author = {Han Wang and Ishwar B Balappanawar and Huan Zhang},
year = {2026},
eprint = {2610.02741},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.02741}
}