Saved in:
| Main Authors: | , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2605.07984 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866918489760989184 |
|---|---|
| author | Ma, Nicole Rui, Nick |
| author_facet | Ma, Nicole Rui, Nick |
| contents | We study planning site formation in language models -- where internal representations of structurally-constrained future tokens form during the forward pass, and whether they causally drive generation. Using rhyming-couplet completion as a clean test of forward-looking constraint, we apply two lightweight methods (linear probing and activation patching) across Qwen3, Gemma-3, and Llama-3 at more than ten scales. Probing shows that future-rhyme information is linearly decodable at the line boundary, with signal that strengthens with scale in all three families. Activation patching reveals that only Gemma-3-27B causally relies on this encoding, exhibiting a handoff in which the causal driver migrates from the rhyme word to the line boundary around layer 30. Every other model we test conditions on the rhyme word throughout generation, with near-zero causal effect at the line boundary despite strong probe signal. We localize the Gemma-3-27B handoff to five attention heads through two-stage path patching that recover ~90% of the rhyme-routing capacity at the newline. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2605_07984 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Where's the Plan? Locating Latent Planning in Language Models with Lightweight Mechanistic Interventions Ma, Nicole Rui, Nick Machine Learning Artificial Intelligence We study planning site formation in language models -- where internal representations of structurally-constrained future tokens form during the forward pass, and whether they causally drive generation. Using rhyming-couplet completion as a clean test of forward-looking constraint, we apply two lightweight methods (linear probing and activation patching) across Qwen3, Gemma-3, and Llama-3 at more than ten scales. Probing shows that future-rhyme information is linearly decodable at the line boundary, with signal that strengthens with scale in all three families. Activation patching reveals that only Gemma-3-27B causally relies on this encoding, exhibiting a handoff in which the causal driver migrates from the rhyme word to the line boundary around layer 30. Every other model we test conditions on the rhyme word throughout generation, with near-zero causal effect at the line boundary despite strong probe signal. We localize the Gemma-3-27B handoff to five attention heads through two-stage path patching that recover ~90% of the rhyme-routing capacity at the newline. |
| title | Where's the Plan? Locating Latent Planning in Language Models with Lightweight Mechanistic Interventions |
| topic | Machine Learning Artificial Intelligence |
| url | https://arxiv.org/abs/2605.07984 |