Paper Citation Record · LEDGER

Understanding Task Vectors in In-Context Learning: Emergence, Functionality, and Limitations

As of 9 August 2026, this Paper Citation Record lists 36 of 36 outbound references and 2 inbound Pith citation observations for arXiv:2506.09048.

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

pith.paper-citation-record.v1
2506.09048 v1

Coverage vector

measured 36 of 36 reference resolution

Typed states for the displayed outbound observations.

Source: paper_references, paper_reference_links, observed 2026-08-07T05:04:25.411130Z

measured 38 of 38 standing notices

One-hop event checks from named stored sources.

Source: scholarly_work_events, retraction_status_cache, observed 2026-08-09T06:31:02.800959+00:00

measured 2 of 2 inbound itemization

Pith citing papers itemized under the disclosed page cap.

Source: paper_references, paper_reference_links, observed 2026-06-29T07:36:53.209887Z

measured 0 of 1 external citation measurements

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

Source: arxiv_reference, observed 2026-06-29T07:43:14.010232Z

Reference resolution

36 of 36 outbound references displayed

verified exact1
verified fuzzy26
unresolved8
parse uncertain0
malformed identifier1
metadata mismatch0

External citation measurements

No source-named external measurement is stored.

Outbound references

Observation 8ca001ee-1d03-489d-b7a2-a0df9af1a287 · outbound

This paper cites Transformers learn to imple- ment preconditioned gradient descent for in-context learning.

Understanding Task Vectors in In-Context Learning: Emergence, Functionality, and Limitations Transformers learn to imple- ment preconditioned gradient descent for in-context learning

Reference 1

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T05:04:25.981043Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-09T06:31:02.800959+00:00.

Observation 46ca5f35-9376-44b6-845f-1c72548dd15c · outbound

This paper cites Language models are few-shot learners.

Understanding Task Vectors in In-Context Learning: Emergence, Functionality, and Limitations Language models are few-shot learners

Reference 2

Resolution: unresolved; no resolver link, observed 2026-08-07T05:04:25.247194Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 272a9053-774f-4e8d-9757-14166634e17f · outbound

This paper cites Data distributional properties drive emer- gent in-context learning in transformers.

Understanding Task Vectors in In-Context Learning: Emergence, Functionality, and Limitations Data distributional properties drive emer- gent in-context learning in transformers

Reference 3

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T05:04:25.954024Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-09T06:31:02.800959+00:00.

Observation e6b16505-fe72-4bd1-93e0-0eee6161fd46 · outbound

This paper cites Why can gpt learn in-context? language models secretly perform gradient descent as meta-optimizers.

Understanding Task Vectors in In-Context Learning: Emergence, Functionality, and Limitations Why can gpt learn in-context? language models secretly perform gradient descent as meta-optimizers

Reference 4

Resolution: unresolved; no resolver link, observed 2026-08-07T05:04:25.256850Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 26300b17-97e4-4cba-a63f-5f794cfe83bc · outbound

This paper cites In-context learning and gradient descent revisited.

Understanding Task Vectors in In-Context Learning: Emergence, Functionality, and Limitations In-context learning and gradient descent revisited

Reference 5

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T05:04:25.927868Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-09T06:31:02.800959+00:00.

Observation 4b568322-e4d4-4de1-9856-ea8c9082c164 · outbound

This paper cites What can transformers learn in-context? a case study of simple function classes.

Understanding Task Vectors in In-Context Learning: Emergence, Functionality, and Limitations What can transformers learn in-context? a case study of simple function classes

Reference 6

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T05:04:25.912255Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-09T06:31:02.800959+00:00.

Observation 54d7dec9-7864-43e5-ba62-14a58dd3f0a3 · outbound

This paper cites In-context learning creates task vectors.

Understanding Task Vectors in In-Context Learning: Emergence, Functionality, and Limitations In-context learning creates task vectors

Reference 7

Resolution: unresolved; no resolver link, observed 2026-08-07T05:04:25.272564Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 87895148-f373-48fe-a071-fd32afb27800 · outbound

This paper cites Finding visual task vectors.

Understanding Task Vectors in In-Context Learning: Emergence, Functionality, and Limitations Finding visual task vectors

Reference 8

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T05:04:25.887670Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-09T06:31:02.800959+00:00.

Observation af9b5e06-5c16-49da-860a-91d91b15e260 · outbound

This paper cites Multimodal task vectors enable many-shot multimodal in-context learning.

Understanding Task Vectors in In-Context Learning: Emergence, Functionality, and Limitations Multimodal task vectors enable many-shot multimodal in-context learning

Reference 9

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T05:04:25.872250Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-09T06:31:02.800959+00:00.

Observation 7259b31f-9d5f-4cbb-a29c-e49634ce2fab · outbound

This paper cites Transformers are rnns: Fast autoregressive transformers with linear attention.

Understanding Task Vectors in In-Context Learning: Emergence, Functionality, and Limitations Transformers are rnns: Fast autoregressive transformers with linear attention

Reference 10

Resolution: unresolved; no resolver link, observed 2026-08-07T05:04:25.286565Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation aa0cfc29-37b2-4707-9db5-a2af96c200a3 · outbound

This paper cites The impact of positional encoding on length generalization in transformers.

Understanding Task Vectors in In-Context Learning: Emergence, Functionality, and Limitations The impact of positional encoding on length generalization in transformers

Reference 11

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T05:04:25.844801Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-09T06:31:02.800959+00:00.

Observation f3a6698f-472c-48d8-bc62-78a0ed5dbe69 · outbound

This paper cites In-context learning state vector with inner and momentum optimization.

Understanding Task Vectors in In-Context Learning: Emergence, Functionality, and Limitations In-context learning state vector with inner and momentum optimization

Reference 12

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T05:04:25.829536Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-09T06:31:02.800959+00:00.

Observation b48f6189-79f2-4a48-9ad6-35de1b2d5c36 · outbound

This paper cites In-context vectors: Making in context learn- ing more effective and controllable through latent space steering.

Understanding Task Vectors in In-Context Learning: Emergence, Functionality, and Limitations In-context vectors: Making in context learn- ing more effective and controllable through latent space steering

Reference 13

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T05:04:25.813203Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-09T06:31:02.800959+00:00.

Observation be412e53-67ea-45e9-b21d-230079c22128 · outbound

This paper cites Mahankali, Tatsunori Hashimoto, and Tengyu Ma.

Understanding Task Vectors in In-Context Learning: Emergence, Functionality, and Limitations Mahankali, Tatsunori Hashimoto, and Tengyu Ma

Reference 14

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T05:04:25.797636Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-09T06:31:02.800959+00:00.

Observation e7861b67-6572-42f1-a522-48fb1d60cc71 · outbound

This paper cites Position: Do pretrained transformers learn in-context by gradient descent? In Proceedings of the 41st International Conference on Machine Learning, pages 44712–44740.

Understanding Task Vectors in In-Context Learning: Emergence, Functionality, and Limitations Position: Do pretrained transformers learn in-context by gradient descent? In Proceedings of the 41st International Conference on Machine Learning, pages 44712–44740

Reference 15

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T05:04:25.781517Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-09T06:31:02.800959+00:00.

Observation a8ec024e-a5f2-4c5d-83d5-3e5b109eb41c · outbound

This paper cites Function vectors in large language models.

Understanding Task Vectors in In-Context Learning: Emergence, Functionality, and Limitations Function vectors in large language models

Reference 16

Resolution: unresolved; no resolver link, observed 2026-08-07T05:04:25.314078Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 0d9d4a57-041d-4bbd-a7ea-7a5da0a5e74d · outbound

This paper cites Transformers learn in-context by gradient descent.

Understanding Task Vectors in In-Context Learning: Emergence, Functionality, and Limitations Transformers learn in-context by gradient descent

Reference 17

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T05:04:25.755114Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-09T06:31:02.800959+00:00.

Observation 0fc1ff61-d8f4-4715-9fd0-1f084a361243 · outbound

This paper cites Label words are anchors: An information flow perspective for understanding in-context learning.

Understanding Task Vectors in In-Context Learning: Emergence, Functionality, and Limitations Label words are anchors: An information flow perspective for understanding in-context learning

Reference 18

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T05:04:25.739972Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-09T06:31:02.800959+00:00.

Observation 2bb39730-8d27-4eae-8e2c-8674fdaeefa9 · outbound

This paper cites On the role of unstructured training data in transformers’ in-context learning capabilities.

Understanding Task Vectors in In-Context Learning: Emergence, Functionality, and Limitations On the role of unstructured training data in transformers’ in-context learning capabilities

Reference 19

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T05:04:25.724578Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-09T06:31:02.800959+00:00.

Observation 04602340-3e7f-4a53-a079-73a67f940744 · outbound

This paper cites How many pretraining tasks are needed for in-context learning of linear regression? In The Twelfth International Conference on Learning Representations , 2024.

Understanding Task Vectors in In-Context Learning: Emergence, Functionality, and Limitations How many pretraining tasks are needed for in-context learning of linear regression? In The Twelfth International Conference on Learning Representations , 2024

Reference 20

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T05:04:25.707827Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-09T06:31:02.800959+00:00.

Observation c0523a64-1144-4980-a799-b9e2dcdb85bb · outbound

This paper cites An explanation of in-context learning as implicit bayesian inference.

Understanding Task Vectors in In-Context Learning: Emergence, Functionality, and Limitations An explanation of in-context learning as implicit bayesian inference

Reference 21

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T05:04:25.692288Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-09T06:31:02.800959+00:00.

Observation 7211e0c5-4688-4d5d-899b-c3eda77eed64 · outbound

This paper cites Theoretical Understanding of In-Context Learning in Shallow Transformers with Unstructured Data.

Understanding Task Vectors in In-Context Learning: Emergence, Functionality, and Limitations Theoretical Understanding of In-Context Learning in Shallow Transformers with Unstructured Data

Reference 22

Resolution: verified exact; local_arxiv, observed 2026-08-07T05:04:25.489160Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-09T06:31:02.800959+00:00.

Observation 9a7ff9a7-71eb-4f40-9861-cd592d0103da · outbound

This paper cites Everything Everywhere All at Once: LLMs can In-Context Learn Multiple Tasks in Superposition.

Understanding Task Vectors in In-Context Learning: Emergence, Functionality, and Limitations Everything Everywhere All at Once: LLMs can In-Context Learn Multiple Tasks in Superposition

Reference 23

Resolution: unresolved; no resolver link, observed 2026-08-07T05:04:25.346396Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 6d173612-cf19-49df-9097-e44561f9a700 · outbound

This paper cites Task Vectors in In-Context Learning: Emergence, Formation, and Benefit.

Understanding Task Vectors in In-Context Learning: Emergence, Functionality, and Limitations Task Vectors in In-Context Learning: Emergence, Formation, and Benefit

Reference 24

Resolution: unresolved; no resolver link, observed 2026-08-07T05:04:25.351904Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation f70e359f-a3ed-41ae-beb9-706ed6c9df9a · outbound

This paper cites Trained transformers learn linear models in-context.

Understanding Task Vectors in In-Context Learning: Emergence, Functionality, and Limitations Trained transformers learn linear models in-context

Reference 25

Resolution: unresolved; no resolver link, observed 2026-08-07T05:04:25.356801Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 66abf830-62f6-4de5-902c-adf2930d1be8 · outbound

This paper cites Position information emerges in causal transformers without positional encodings via similarity of nearby embeddings.

Understanding Task Vectors in In-Context Learning: Emergence, Functionality, and Limitations Position information emerges in causal transformers without positional encodings via similarity of nearby embeddings

Reference 26

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T05:04:25.666527Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-09T06:31:02.800959+00:00.

Observation d23eea73-4911-4693-a510-8ed26ae62f1e · outbound

This paper cites We first show that for any l ∈ [1, L], the following equations hold: Xl(X0 × ← −UΣ) = UΣXl, (16) d dt Xl(X0 × ← −UΣ, Ai + ← −tR) t=0 = UΣ d dt Xl(Ai + ← −tU −1 Σ RUΣ) t=0.

Understanding Task Vectors in In-Context Learning: Emergence, Functionality, and Limitations We first show that for any l ∈ [1, L], the following equations hold: Xl(X0 × ← −UΣ) = UΣXl, (16) d dt Xl(X0 × ← −UΣ, Ai + ← −tR) t=0 = UΣ d dt Xl(Ai + ← −tU −1 Σ RUΣ) t=0

Reference 27

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T05:04:25.650955Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-09T06:31:02.800959+00:00.

Observation 84226d19-440e-4d1c-bcea-455761451586 · outbound

This paper cites From the recursive expressions in eq.

Understanding Task Vectors in In-Context Learning: Emergence, Functionality, and Limitations From the recursive expressions in eq

Reference 28

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T05:04:25.635253Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-09T06:31:02.800959+00:00.

Observation f43163e7-a8a5-4ff9-a9a6-8619058cd50f · outbound

This paper cites Similar to the Ai case, we will first prove that for any l ∈ [1, L], d dt Xl(X0 × ← −UΣ, Ci + ← −tR) t=0 = UΣ d dt Xl(Ci + ← −tU ⊤ Σ RUΣ) t=0.

Understanding Task Vectors in In-Context Learning: Emergence, Functionality, and Limitations Similar to the Ai case, we will first prove that for any l ∈ [1, L], d dt Xl(X0 × ← −UΣ, Ci + ← −tR) t=0 = UΣ d dt Xl(Ci + ← −tU ⊤ Σ RUΣ) t=0

Reference 29

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T05:04:25.619535Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-09T06:31:02.800959+00:00.

Observation 54894bdd-c6e9-47d9-b08a-776a314a76bb · outbound

This paper cites Let Up ∈ Rn×n be a uniformly sampled permutation matrix, i.e., a binary matrix that has exactly one1 entry in each row and column with all other entries0.

Understanding Task Vectors in In-Context Learning: Emergence, Functionality, and Limitations Let Up ∈ Rn×n be a uniformly sampled permutation matrix, i.e., a binary matrix that has exactly one1 entry in each row and column with all other entries0

Reference 30

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T05:04:25.603904Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-09T06:31:02.800959+00:00.

Observation 8943c6da-6259-4be7-a86b-e017b746afa3 · outbound

This paper cites 21 One can easily verify that eqs.

Understanding Task Vectors in In-Context Learning: Emergence, Functionality, and Limitations 21 One can easily verify that eqs

Reference 31

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T05:04:25.588296Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-09T06:31:02.800959+00:00.

Observation 894f59e1-1e83-4156-af59-47908006c340 · outbound

This paper cites From the definition of Fl and Gl, we can verify that d dt Yl(Bi + ← −tR) t=0 = R(Fi−1 + W Gi−1)M (X ⊤ i−1CiXi−1 + Di) lY j=i+1 I + bjM (X ⊤ j−1CjXj−1 + Dj).

Understanding Task Vectors in In-Context Learning: Emergence, Functionality, and Limitations From the definition of Fl and Gl, we can verify that d dt Yl(Bi + ← −tR) t=0 = R(Fi−1 + W Gi−1)M (X ⊤ i−1CiXi−1 + Di) lY j=i+1 I + bjM (X ⊤ j−1CjXj−1 + Dj)

Reference 32

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T05:04:25.570755Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-09T06:31:02.800959+00:00.

Observation cd0f089b-e569-4ea1-b72e-ff05a24af687 · outbound

This paper cites Similarly, one can verify that eqs.

Understanding Task Vectors in In-Context Learning: Emergence, Functionality, and Limitations Similarly, one can verify that eqs

Reference 33

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T05:04:25.554695Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-09T06:31:02.800959+00:00.

Observation 0cb6a197-6ef5-4386-a7ea-606b9fbdd7e6 · outbound

This paper cites Similarly, we have Y1 = Y0 + b1Y0M diag(In ⊗ D1 1, D2.

Understanding Task Vectors in In-Context Learning: Emergence, Functionality, and Limitations Similarly, we have Y1 = Y0 + b1Y0M diag(In ⊗ D1 1, D2

Reference 34

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T05:04:25.538021Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-09T06:31:02.800959+00:00.

Observation e8407f5b-245c-4593-a436-1085fdfa6ea7 · outbound

This paper cites By the definition of linear attention, we can show that TF(Z0; {Vl, Ql}2 l=1) = (Y2)3n+3 = b2Y1M c2X ⊤ 1 (X1)3n+3 + (D2)3n+3 = b2c2a1dy x 3n+2X i=1 (Y1)i(X1)⊤ i ! xtest.

Understanding Task Vectors in In-Context Learning: Emergence, Functionality, and Limitations By the definition of linear attention, we can show that TF(Z0; {Vl, Ql}2 l=1) = (Y2)3n+3 = b2Y1M c2X ⊤ 1 (X1)3n+3 + (D2)3n+3 = b2c2a1dy x 3n+2X i=1 (Y1)i(X1)⊤ i ! xtest

Reference 35

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T05:04:25.521435Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-09T06:31:02.800959+00:00.

Observation 4e16e78c-648e-4a7e-9e62-7b187b4da335 · outbound

This paper cites For training efficiency and stability, we restrict theAl, Bl, and Cl matrices to SI during training, and initialize Dl ∈ Rdp×dp with i.i.d.

Understanding Task Vectors in In-Context Learning: Emergence, Functionality, and Limitations For training efficiency and stability, we restrict theAl, Bl, and Cl matrices to SI during training, and initialize Dl ∈ Rdp×dp with i.i.d

Reference 1000

Resolution: malformed identifier; raw_fallback, observed 2026-08-07T05:04:25.505494Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-09T06:31:02.800959+00:00.

Pith citing papers

Observation 5282e0a2-7cf0-49c0-a0c1-3044653d76a9 · inbound

Distributional Alignment as a Criterion for Designing Task Vectors in In-Context Learning cites this paper.

Distributional Alignment as a Criterion for Designing Task Vectors in In-Context Learning Understanding Task Vectors in In-Context Learning: Emergence, Functionality, and Limitations

Reference 10

Resolution: verified exact; arxiv_id, observed 2026-05-21T05:33:58.620574Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-09T06:31:02.800959+00:00.

Observation 38e228fb-e62f-4cb3-9a0f-4330647a39e8 · inbound

Causal Interventions on Continuous Variables: A Case Study on Verb Bias in Steering Vectors for In-Context Learning cites this paper.

Causal Interventions on Continuous Variables: A Case Study on Verb Bias in Steering Vectors for In-Context Learning Understanding Task Vectors in In-Context Learning: Emergence, Functionality, and Limitations

Reference 12

Resolution: verified exact; arxiv_id, observed 2026-06-29T07:43:14.011735Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-09T06:31:02.800959+00:00.