A Shared Subcircuit Lets LLMs Count Down Across Tasks
Quick Answer
The study identifies a 'countdown subcircuit' in Llama-3.1-70B-Instruct, enabling language models to track token counts across various tasks.
Quick Take
The study identifies a 'countdown subcircuit' in Llama-3.1-70B-Instruct, enabling language models to track token counts across various tasks. This mechanism, shared among models, enhances performance in tasks like sentence writing and DNA sequence formatting, suggesting broader implications for understanding model behavior generalization.
Key Points
- The countdown subcircuit helps models complete tasks requiring token tracking.
- Identified in Llama-3.1-70B-Instruct, it shows shared motifs across models.
- Tasks include writing fixed-length sentences and formatting ASCII tables.
- Unsupervised probing reveals additional tasks utilizing this subcircuit.
- Understanding subcircuits aids in generalizing model behaviors across tasks.
Paper Resources
📖 Reader Mode
~2 min readAbstract:Writing a sentence of exactly twelve words; ending a DNA sequence at the right codon; formatting an ASCII table. These are all tasks that language models can do that requires tracking how many tokens remain before a target. In this work, we identify in Llama-3.1-70B-Instruct a general mechanism for performing these tasks: a "countdown subcircuit" that compares the current position to a goal length and estimates the time remaining until then. We first isolate a countdown subcircuit in a controlled setting, in which the model is tasked with writing a fixed-length sentence ending in a specified word. We then investigate the geometry of the representations used by the subcircuit, and find that the subcircuit uses an identical motif previously identified in a frontier LLM on a separate task, thus suggesting that this motif is shared across models. Finally, we use unsupervised probing on a natural language dataset to find a variety of other tasks where this subcircuit is used, including tasks where the goal length is inferred from context rather than explicitly stated. Our work suggests that reverse-engineering subcircuits allows us to understand how behaviors generalize from a single example to many different tasks and even models.
| Comments: | 12 pages, 11 figures |
| Subjects: | Computation and Language (cs.CL); Machine Learning (cs.LG) |
| Cite as: | arXiv:2607.12279 [cs.CL] |
| (or arXiv:2607.12279v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2607.12279 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Jacob Dunefsky [view email]
[v1]
Tue, 14 Jul 2026 02:30:45 UTC (1,437 KB)
— Originally published at arxiv.org
Want this in your inbox every morning?
Daily brief at your local 8am — bilingual EN/中文, free.
More from arXiv cs.CL
See more →Letting the Data Speak: Extracting Keywords from Crowdsourced Collections with AI
The study evaluates three NLP approaches—Named Entity Recognition, Keyword Extraction, and Topic Modelling—using the Their Finest Hour Online Archive to automate keyword extraction from crowdsourced WWII collections. Findings suggest that while NLP methods show promise, no single approach is sufficient, and ethical considerations in automated keyword extraction are crucial for responsible stewardship.