Pith. sign in

Paper Citation Record · LEDGER

What Makes Position Zero Special? A Mechanistic Study of Position Zero Attention Sinks in LLMs

As of 6 August 2026, this Paper Citation Record lists 21 of 21 outbound references and 2 inbound Pith citation observations for arXiv:2603.06591.

A citation records a reference. It does not transfer a finding from one paper to another.

pith.paper-citation-record.v1
2603.06591 v2

Coverage vector

measured 21 of 21 reference resolution

Typed states for the displayed outbound observations.

Source: paper_references, paper_reference_links, observed 2026-08-04T06:15:45.979625Z

measured 23 of 23 standing notices

One-hop event checks from named stored sources.

Source: scholarly_work_events, retraction_status_cache, observed 2026-08-05T06:32:48.257954+00:00

measured 2 of 2 inbound itemization

Pith citing papers itemized under the disclosed page cap.

Source: paper_references, paper_reference_links, observed 2026-06-28T15:11:32.527978Z

measured 0 of 1 external citation measurements

A source-named dated measurement, never combined with another source.

Source: arxiv_reference, observed 2026-07-01T22:36:17.704896Z

Reference resolution

21 of 21 outbound references displayed

  • verified exact0
  • verified fuzzy0
  • unresolved20
  • parse uncertain0
  • malformed identifier1
  • metadata mismatch0

External citation measurements

No source-named external measurement is stored.

Outbound references

Observation fd7e7498-130d-4c9a-aea5-7ab74a0f6443 · outbound

This paper cites Layer Normalization.

What Makes Position Zero Special? A Mechanistic Study of Position Zero Attention Sinks in LLMs Layer Normalization

Reference 1

Resolution
unresolved
no resolver link, observed 2026-08-04T06:15:44.732896Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-04T06:15:44.732896Z digest=sha256:2bf80f1699d2a623ac25f1d0d9d9f26420a4c15e18fd9f1b839f69e590a19ba0

Observation b9cfd999-ec27-4e31-9c98-ff992c29be22 · outbound

This paper cites Why do LLMs attend to the first token?.

What Makes Position Zero Special? A Mechanistic Study of Position Zero Attention Sinks in LLMs Why do LLMs attend to the first token?

Reference 2

Resolution
unresolved
no resolver link, observed 2026-08-04T06:15:44.817870Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-04T06:15:44.817870Z digest=sha256:3592c9f75ad9afe0f2d8968fb9b7be328680e92bb99e85a64860853e50366df6

Observation 762341ed-8f2a-427c-9098-19b871c537cc · outbound

This paper cites Theoretical Foundations of t-SNE for Visualizing High-Dimensional Clustered Data.

What Makes Position Zero Special? A Mechanistic Study of Position Zero Attention Sinks in LLMs Theoretical Foundations of t-SNE for Visualizing High-Dimensional Clustered Data

Reference 3

Resolution
unresolved
no resolver link, observed 2026-08-04T06:15:44.875017Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-04T06:15:44.875017Z digest=sha256:65881c3f953580d7974c8475535b0563c3162a092b8c343f8e8a1f79ade413f7

Observation 03a2aa62-12ee-46cb-994e-2bf7bc33c9d0 · outbound

This paper cites Spectral Filters, Dark Signals, and Attention Sinks.

What Makes Position Zero Special? A Mechanistic Study of Position Zero Attention Sinks in LLMs Spectral Filters, Dark Signals, and Attention Sinks

Reference 4

Resolution
unresolved
no resolver link, observed 2026-08-04T06:15:44.927705Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-04T06:15:44.927705Z digest=sha256:ca7d6c51528b304b788a183dbdb91556db188982be5c64db993f47b25661f6dd

Observation 95f26965-95d3-4d02-8ee9-4782fd95b59f · outbound

This paper cites MagicPIG: LSH Sampling for Efficient LLM Generation.

What Makes Position Zero Special? A Mechanistic Study of Position Zero Attention Sinks in LLMs MagicPIG: LSH Sampling for Efficient LLM Generation

Reference 5

Resolution
unresolved
no resolver link, observed 2026-08-04T06:15:44.984092Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-04T06:15:44.984092Z digest=sha256:e0828645db787a2e639ce1789737ed194791a17df3aedb7886c5c261f3d50bdc

Observation a245cc30-7c9c-4ce0-af70-8c566836a620 · outbound

This paper cites The emergence of clusters in self-attention dynamics.

What Makes Position Zero Special? A Mechanistic Study of Position Zero Attention Sinks in LLMs The emergence of clusters in self-attention dynamics

Reference 6

Resolution
unresolved
no resolver link, observed 2026-08-04T06:15:45.055088Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-04T06:15:45.055088Z digest=sha256:d643663719ab099709a499a0df6846ad8d5db2af356cb908d7c972fe7363488b

Observation a6be503c-9e00-4604-a30a-8ceedd8e70c8 · outbound

This paper cites The Llama 3 Herd of Models.

What Makes Position Zero Special? A Mechanistic Study of Position Zero Attention Sinks in LLMs The Llama 3 Herd of Models

Reference 7

Resolution
unresolved
no resolver link, observed 2026-08-04T06:15:45.149007Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-04T06:15:45.149007Z digest=sha256:14134550c3062262e12ddfe48a6966578b3653a09f3126d0a2c0934d503f5cfe

Observation 519b3ead-c985-4fe2-8fbf-21b07849d65d · outbound

This paper cites When Attention Sink Emerges in Language Models: An Empirical View.

What Makes Position Zero Special? A Mechanistic Study of Position Zero Attention Sinks in LLMs When Attention Sink Emerges in Language Models: An Empirical View

Reference 8

Resolution
unresolved
no resolver link, observed 2026-08-04T06:15:45.217912Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-04T06:15:45.217912Z digest=sha256:8371fb7c87d1cac80d5b336cf9ef6881049c13c42595a780e85eb342c92ddd1b

Observation 074b3f0a-b670-421c-9106-442ef0c1f6b4 · outbound

This paper cites Active-dormantattentionheads: Mechanistically demystifying extreme-token phenomena in LLMs, 2025.

What Makes Position Zero Special? A Mechanistic Study of Position Zero Attention Sinks in LLMs Active-dormantattentionheads: Mechanistically demystifying extreme-token phenomena in LLMs, 2025

Reference 9

Resolution
unresolved
no resolver link, observed 2026-08-04T06:15:45.257668Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-04T06:15:45.257668Z digest=sha256:efb750ebce5357d13f6f022f073efe23f24a4da717a7fa6063fcd3859880217b

Observation c8c00357-2719-4c54-993d-619311cba654 · outbound

This paper cites LM-infinite: Zero-shot extreme length generalization for large language models.

What Makes Position Zero Special? A Mechanistic Study of Position Zero Attention Sinks in LLMs LM-infinite: Zero-shot extreme length generalization for large language models

Reference 10

Resolution
unresolved
no resolver link, observed 2026-08-04T06:15:45.302928Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-04T06:15:45.302928Z digest=sha256:21f4d11a650eb8e1e66efc9beb4b63c45596d7b5c77dde156701092c5001e2f7

Observation 9d59e1c6-2595-4de4-b1c1-e47c593fb804 · outbound

This paper cites Fineweb-edu: the finest col- lection of educational content, 2024.

What Makes Position Zero Special? A Mechanistic Study of Position Zero Attention Sinks in LLMs Fineweb-edu: the finest col- lection of educational content, 2024

Reference 11

Resolution
unresolved
no resolver link, observed 2026-08-04T06:15:45.356911Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-04T06:15:45.356911Z digest=sha256:5f106bb7af9d388c25f4acb17765b615fdd30fdb29f1909661406051aa928f2c

Observation 9fe0327a-bece-40e5-be7e-09642874b77e · outbound

This paper cites 2 OLMo 2 Furious.

What Makes Position Zero Special? A Mechanistic Study of Position Zero Attention Sinks in LLMs 2 OLMo 2 Furious

Reference 12

Resolution
unresolved
no resolver link, observed 2026-08-04T06:15:45.440129Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-04T06:15:45.440129Z digest=sha256:5afd813e6117db448dd357c26aaa13ee7a2d9dc8faf23b0907ab48cc2288af1d

Observation c95e9c1b-181b-46d4-abed-afa0b43fd4eb · outbound

This paper cites WanJuan-CC: A Safe and High-Quality Open-sourced English Webtext Dataset.

What Makes Position Zero Special? A Mechanistic Study of Position Zero Attention Sinks in LLMs WanJuan-CC: A Safe and High-Quality Open-sourced English Webtext Dataset

Reference 13

Resolution
unresolved
no resolver link, observed 2026-08-04T06:15:45.517494Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-04T06:15:45.517494Z digest=sha256:c5fb223f1c0555f842a2ff6daba223eadd849d657bde287d9fd58a544a07f7a9

Observation e07b8d76-666c-4405-9032-0873d054ea89 · outbound

This paper cites Qwen2.5 Technical Report.

What Makes Position Zero Special? A Mechanistic Study of Position Zero Attention Sinks in LLMs Qwen2.5 Technical Report

Reference 14

Resolution
unresolved
no resolver link, observed 2026-08-04T06:15:45.576945Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-04T06:15:45.576945Z digest=sha256:9f0579721892f802071ba710771ff4cb4fe61ca51e50cb501b306f258a3bd867

Observation 5183aab2-a868-4d0e-afb4-e351805dedd1 · outbound

This paper cites Identifying and evaluating inactive heads in pretrained llms, 2025.

What Makes Position Zero Special? A Mechanistic Study of Position Zero Attention Sinks in LLMs Identifying and evaluating inactive heads in pretrained llms, 2025

Reference 15

Resolution
unresolved
no resolver link, observed 2026-08-04T06:15:45.634756Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-04T06:15:45.634756Z digest=sha256:e89d0b5b828eb41fbb6af2d65f0fbb8e32287e3a6f5626cfcde7fe9fc72a8c06

Observation aa3e50ad-6be5-4ea3-9af8-c9ed592a114f · outbound

This paper cites Efficient Streaming Language Models with Attention Sinks.

What Makes Position Zero Special? A Mechanistic Study of Position Zero Attention Sinks in LLMs Efficient Streaming Language Models with Attention Sinks

Reference 16

Resolution
unresolved
no resolver link, observed 2026-08-04T06:15:45.689712Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-04T06:15:45.689712Z digest=sha256:5741146a42d87f5caa2a83b7aff80b18d60b02349af350cd9f3de5f344054d4a

Observation 930dad20-5b3d-4135-91f5-413f543860ef · outbound

This paper cites Qwen3 Technical Report.

What Makes Position Zero Special? A Mechanistic Study of Position Zero Attention Sinks in LLMs Qwen3 Technical Report

Reference 17

Resolution
unresolved
no resolver link, observed 2026-08-04T06:15:45.754739Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-04T06:15:45.754739Z digest=sha256:c92c06b6b4000e55339ef71e07a34fc4ad4808f4444fa0143e79b1d74ded19de

Observation ca10b16b-864f-4190-a90d-f7279d962d68 · outbound

This paper cites Unveiling and Harnessing Hidden Attention Sinks: Enhancing Large Language Models without Training through Attention Calibration.

What Makes Position Zero Special? A Mechanistic Study of Position Zero Attention Sinks in LLMs Unveiling and Harnessing Hidden Attention Sinks: Enhancing Large Language Models without Training through Attention Calibration

Reference 18

Resolution
unresolved
no resolver link, observed 2026-08-04T06:15:45.793248Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-04T06:15:45.793248Z digest=sha256:76594147bed92246bb3543145990be4aeadc0d32d72578cfeb5c493106324750

Observation 746e3b3d-c37e-4297-86c2-74741dffa449 · outbound

This paper cites Root Mean Square Layer Normalization.

What Makes Position Zero Special? A Mechanistic Study of Position Zero Attention Sinks in LLMs Root Mean Square Layer Normalization

Reference 19

Resolution
unresolved
no resolver link, observed 2026-08-04T06:15:45.856712Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-04T06:15:45.856712Z digest=sha256:b310e54b585618b1683d35ce9971fd5213c7c74cccfc270f0f0c3f931f703502

Observation 21b8f318-4d42-4153-bd9c-794726c953ee · outbound

This paper cites Attention sinks: A ’catch, tag, release’ mechanism for embeddings, 2025.

What Makes Position Zero Special? A Mechanistic Study of Position Zero Attention Sinks in LLMs Attention sinks: A ’catch, tag, release’ mechanism for embeddings, 2025

Reference 20

Resolution
unresolved
no resolver link, observed 2026-08-04T06:15:45.926232Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-04T06:15:45.926232Z digest=sha256:642e34735b2164feed0ec37099aa53d62ba1360ba56e1da2592aeef307f29c4a

Observation cac063aa-6701-4bb0-9b17-12be72649108 · outbound

This paper cites an unresolved cited work.

What Makes Position Zero Special? A Mechanistic Study of Position Zero Attention Sinks in LLMs Unresolved cited work

Reference 21

Resolution
malformed identifier
no resolver link, observed 2026-08-04T06:15:45.979625Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

source=pdf_text observed=2026-08-04T06:15:45.979625Z digest=sha256:001374519b2b2efb90e310bea3d8ca1405b7786a247d3de27d5eb12bba937fa5

Pith citing papers

Observation f487e507-5162-4eee-9b2e-1a3d74c828fe · inbound

Irminsul: MLA-Native Position-Independent Caching for Agentic LLM Serving cites this paper.

Irminsul: MLA-Native Position-Independent Caching for Agentic LLM Serving What Makes Position Zero Special? A Mechanistic Study of Position Zero Attention Sinks in LLMs

Reference 26

Resolution
verified exact
arxiv_id, observed 2026-08-04T02:30:25.129019Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

source=pdf_text observed=2026-05-08T05:37:02.419657Z digest=sha256:aa3cb3e891535790de3b3f9273408e0b779eece731dde75b2281bb0fe8672ef0

Observation af4e6a35-f806-41cf-ac8c-3254e877880e · inbound

From Zero to Hero: Training-Free Custom Concept Spawning in World Models cites this paper.

From Zero to Hero: Training-Free Custom Concept Spawning in World Models What Makes Position Zero Special? A Mechanistic Study of Position Zero Attention Sinks in LLMs

Reference 24

Resolution
metadata mismatch
arxiv_id, observed 2026-08-04T02:30:25.129019Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-05T06:32:48.257954+00:00.

source=pdf_text observed=2026-06-28T15:11:32.527978Z digest=sha256:65a444db9db4cc61777120c35a49dcb7b1623ea305929454f592472aa428a72d