IAAIInternalizing CoT into Loops
Internalizing Chain-of-thought into Looped Transformers
A theoretical construction of constant-bit-size looped transformers that interpolates between chain-of-thought and latent recurrence, trading generated tokens for loop iterations when simulating Turing machines.
Inside the method
Looped transformer with L loops + reduced CoT tokens
Simplified conceptual schematic. Consult the paper for the complete architecture.
- Recurrence family
- Full-stack recurrence
- Depth control
- Prescribed loop count L(n)
- KV / state strategy
- Context window O(s(n)/L)
Reading note
Reduces both context length and generated tokens by a factor of L relative to the unlooped queue-machine simulation, including with softmax attention at constant bit size. Theory only; no trained models.
Sources checked 2026-10-28. This catalog does not imply independent reproduction.
Related works
Cite this work
@misc{cottolooped2026,
title = {Internalizing Chain-of-thought into Looped Transformers},
author = {Kehang Zhou and Pinyan Lu},
year = {2026},
eprint = {2610.25998},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.25998}
}