Gathered, Not Admitted: How Attention Brings a Latent Variable into Verbalizable Form
Summary
This paper explores how attention mechanisms in language models render latent variables accessible without a selective gate, identifying a demand-specific mid-depth window where attention-mediated gathering occurs.
View Cached Full Text
Cached at: 08/18/26, 07:53 AM
Paper page - Gathered, Not Admitted: How Attention Brings a Latent Variable into Verbalizable Form
Source: https://huggingface.co/papers/2608.15022
Abstract
In language models, flexible reuse demands attention-mediated gathering at a mid-depth window to make latent variables readable, without a selective gate, and readout measures poorly reflect actual use.
Language models hold latent quantities in a form they can report on, and more of a quantity is present in that form when the task requires reusing it flexibly. What causes a representation to enter that form is open, and the wordworkspaceinvites an admission story: a gate that decides what gets in. Testing it onopen-weight modelswithJacobian lenses, over a benchmark whose five arms share an identical context, we find no gate where it predicts one. Demand raises a concept’s lens visibility beyond what applying an operator to a supplied value produces: +0.050 [+0.045, +0.057] inpercentile rankon our primary checkpoint, positive on all four we measure, though that arm answers at ceiling and the accuracymatched contrast is stronger under thatreadout. At the same time one sharedlinear mapdecodes the variable from every arm, the control included, at 6.4-9.0x its selection-corrected floor. What produces the later readable form at the queried position isattention-mediated gatheringinside amid-depth window: separatingpatch depthfromreadoutdepth puts transport there at least 17x above anywhere shallower under non-saturatingreadouts, with no testedMLPoutput contributing positively inside it. Under the saturatingpercentile rankthe same grid does not localise the window, which is a fact about that measure. An arm that needs the variable for nothing concentrates sevenfold less, so the window is demand-specific. That window has two measured edges, a survival failure below and destruction above, and it falls at the same fractional depth in a 64-layer hybrid and a 62-layerdense modelfrom another family. We localise where the variable is installed and read, not the route from the passage, which transports nothing. But thereadoutis not a calibrated measure of use: three components move it to within 12% of one another and differ 7.4x in what they do to the answer.
View arXiv pageView PDFProject pageGitHub0Add to collection
Get this paper in your agent:
hf papers read 2608\.15022
Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash
Models citing this paper0
No model linking this paper
Cite arxiv.org/abs/2608.15022 in a model README.md to link it from this page.
Datasets citing this paper0
No dataset linking this paper
Cite arxiv.org/abs/2608.15022 in a dataset README.md to link it from this page.
Spaces citing this paper0
No Space linking this paper
Cite arxiv.org/abs/2608.15022 in a Space README.md to link it from this page.
Collections including this paper0
No Collection including this paper
Add this paper to acollectionto link it from this page.
Similar Articles
Rethinking the Role of Efficient Attention in Hybrid Architectures
This paper systematically analyzes the role of efficient attention modules in hybrid language model architectures, finding that different designs converge in long-context performance under sufficient training, and that long-range retrieval is primarily carried by full attention while efficient attention shapes the optimization trajectory, revealing a 'Large-Window Laziness' phenomenon.
Attention Degradation, Function Token Anchoring, and the Limits of Attention-Based Intervention in Large Language Models
This paper investigates short-term attention degradation in LLMs, finding a universal exponential-then-plateau pattern and that function token anchoring is architecture-dependent. Causal tests show that increasing attention mass on function tokens does not improve retrieval, suggesting attention degradation is descriptive rather than prescriptive.
The Structural Attention Tax: How Retrieval Format Hijacks In-Context Learning Independent of Content
This paper identifies and formalizes the 'structural attention tax' phenomenon, where the format of retrieved content (e.g., knowledge graph triples) independently distorts LLM attention distribution regardless of semantic relevance, leading to compressed demonstration attention. It provides a formal framework, empirical evidence across models and benchmarks, and proposes structure-aware mitigation strategies.
@shikhargupta02: I’ve been learning about latent attention (by deepseek). Instead of storing a full K and a V vector per token, it rathe…
The author shares insights from training a small model with DeepSeek's latent attention, observing layer-dependent latent usage and a test-time trick that reduces KV cache 4x without loss change.
Where Reliability Lives in Vision-Language Models: A Mechanistic Study of Attention, Hidden States, and Causal Circuits
This paper challenges the 'Attention-Confidence Assumption' by demonstrating that attention map sharpness is a poor predictor of correctness in Vision-Language Models. Instead, it shows that reliability is better indicated by hidden-state geometry and self-consistency, with significant findings on architectural differences between late-fusion and early-fusion models.