Paper Citation Record · LEDGER

Understanding Fact Recall in Language Models: Why Two-Stage Training Encourages Memorization but Mixed Training Teaches Knowledge

As of 10 August 2026, this Paper Citation Record lists 54 of 54 outbound references and 3 inbound Pith citation observations for arXiv:2505.16178.

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

pith.paper-citation-record.v1
2505.16178 v2

Coverage vector

measured 54 of 54 reference resolution

Typed states for the displayed outbound observations.

Source: paper_references, paper_reference_links, observed 2026-08-07T15:12:00.573379Z

measured 57 of 57 standing notices

One-hop event checks from named stored sources.

Source: scholarly_work_events, retraction_status_cache, observed 2026-08-10T06:31:04.303077+00:00

measured 3 of 3 inbound itemization

Pith citing papers itemized under the disclosed page cap.

Source: paper_references, paper_reference_links, observed 2026-08-05T20:53:40.719452Z

measured 0 of 1 external citation measurements

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

Source: pith, observed 2026-07-04T14:19:53.912897Z

Reference resolution

54 of 54 outbound references displayed

verified exact0
verified fuzzy44
unresolved9
parse uncertain0
malformed identifier1
metadata mismatch0

External citation measurements

No source-named external measurement is stored.

Outbound references

Observation b3b0063e-b53a-4442-9ee6-3de35ee46ad9 · outbound

This paper cites Evaluating correctness and faithfulness of instruction-following models for question answering.

Understanding Fact Recall in Language Models: Why Two-Stage Training Encourages Memorization but Mixed Training Teaches Knowledge Evaluating correctness and faithfulness of instruction-following models for question answering

Reference 1

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T15:12:02.909216Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 9cc26fe9-fad4-4789-b343-d91b52fa66f7 · outbound

This paper cites Llama 3.2: Revolutionizing edge ai and vision with open, customizable models.

Understanding Fact Recall in Language Models: Why Two-Stage Training Encourages Memorization but Mixed Training Teaches Knowledge Llama 3.2: Revolutionizing edge ai and vision with open, customizable models

Reference 2

Resolution: unresolved; no resolver link, observed 2026-08-07T15:11:55.475764Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation e41ee26d-26d9-4ee4-a093-efb3b83ee40f · outbound

This paper cites Physics of language models: Part 3.1, knowledge storage and extraction.

Understanding Fact Recall in Language Models: Why Two-Stage Training Encourages Memorization but Mixed Training Teaches Knowledge Physics of language models: Part 3.1, knowledge storage and extraction

Reference 3

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T15:12:02.881252Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 13869605-8315-40d7-b021-776c72b2b662 · outbound

This paper cites Physics of language models: Part 3.2, knowledge manipula- tion.

Understanding Fact Recall in Language Models: Why Two-Stage Training Encourages Memorization but Mixed Training Teaches Knowledge Physics of language models: Part 3.2, knowledge manipula- tion

Reference 4

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T15:12:02.865284Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 59b507d6-e94e-4515-be59-e36ee589cdf8 · outbound

This paper cites Towards better understanding of gradient-based attribution methods for deep neural networks.

Understanding Fact Recall in Language Models: Why Two-Stage Training Encourages Memorization but Mixed Training Teaches Knowledge Towards better understanding of gradient-based attribution methods for deep neural networks

Reference 5

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T15:12:02.849487Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation bb00f3c9-52e9-4000-9e30-a8303963cfb8 · outbound

This paper cites Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback.

Understanding Fact Recall in Language Models: Why Two-Stage Training Encourages Memorization but Mixed Training Teaches Knowledge Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback

Reference 6

Resolution: unresolved; no resolver link, observed 2026-08-07T15:11:55.947350Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 2d020124-f740-4d2c-9053-8bb795e7911c · outbound

This paper cites Pythia: A suite for analyzing large language models across training and scaling.

Understanding Fact Recall in Language Models: Why Two-Stage Training Encourages Memorization but Mixed Training Teaches Knowledge Pythia: A suite for analyzing large language models across training and scaling

Reference 7

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T15:12:02.833474Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 195992c6-268c-442d-8397-2f86538a3eff · outbound

This paper cites Instructpix2pix: Learning to follow image editing instructions.

Understanding Fact Recall in Language Models: Why Two-Stage Training Encourages Memorization but Mixed Training Teaches Knowledge Instructpix2pix: Learning to follow image editing instructions

Reference 8

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T15:12:02.817716Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 2360e71c-2c40-47e3-ae0a-0bc67a629fff · outbound

This paper cites Causal scrubbing: A method for rigorously testing interpretability hypotheses.

Understanding Fact Recall in Language Models: Why Two-Stage Training Encourages Memorization but Mixed Training Teaches Knowledge Causal scrubbing: A method for rigorously testing interpretability hypotheses

Reference 9

Resolution: unresolved; no resolver link, observed 2026-08-07T15:11:56.397434Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 6a5d02bf-22bf-4698-b299-fe3e35741a63 · outbound

This paper cites Journey to the center of the knowledge neurons: Discoveries of language-independent knowledge neurons and degenerate knowledge neurons.

Understanding Fact Recall in Language Models: Why Two-Stage Training Encourages Memorization but Mixed Training Teaches Knowledge Journey to the center of the knowledge neurons: Discoveries of language-independent knowledge neurons and degenerate knowledge neurons

Reference 10

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T15:12:02.790929Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 8d2b9e9e-e811-48bf-9e9b-738bea0e7407 · outbound

This paper cites Instruc- tion pre-training: Language models are supervised multitask learners.

Understanding Fact Recall in Language Models: Why Two-Stage Training Encourages Memorization but Mixed Training Teaches Knowledge Instruc- tion pre-training: Language models are supervised multitask learners

Reference 11

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T15:12:02.772342Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 4c73c785-e1a6-4ccb-a084-1ae57e33ea00 · outbound

This paper cites Zhao, Yanping Huang, Andrew M.

Understanding Fact Recall in Language Models: Why Two-Stage Training Encourages Memorization but Mixed Training Teaches Knowledge Zhao, Yanping Huang, Andrew M

Reference 12

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T15:12:02.756083Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation dd64acd9-f7d6-45cb-aa59-4472c17c6751 · outbound

This paper cites Towards automated circuit discovery for mechanistic interpretability.

Understanding Fact Recall in Language Models: Why Two-Stage Training Encourages Memorization but Mixed Training Teaches Knowledge Towards automated circuit discovery for mechanistic interpretability

Reference 13

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T15:12:02.740201Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 7b27cf0d-cc75-4629-9085-66d95b2803c0 · outbound

This paper cites Free dolly: Introducing the world’s first truly open instruction-tuned llm, 2023.

Understanding Fact Recall in Language Models: Why Two-Stage Training Encourages Memorization but Mixed Training Teaches Knowledge Free dolly: Introducing the world’s first truly open instruction-tuned llm, 2023

Reference 14

Resolution: unresolved; no resolver link, observed 2026-08-07T15:11:56.908529Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 9897528a-d134-4077-9dbb-06064f5b0d74 · outbound

This paper cites Knowledge neurons in pretrained transformers.

Understanding Fact Recall in Language Models: Why Two-Stage Training Encourages Memorization but Mixed Training Teaches Knowledge Knowledge neurons in pretrained transformers

Reference 15

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T15:12:02.712936Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 4554704a-1992-437f-993f-6e83a25c1ccf · outbound

This paper cites InstructBLIP: Towards general-purpose vision-language models with instruction tuning.

Understanding Fact Recall in Language Models: Why Two-Stage Training Encourages Memorization but Mixed Training Teaches Knowledge InstructBLIP: Towards general-purpose vision-language models with instruction tuning

Reference 16

Resolution: unresolved; no resolver link, observed 2026-08-07T15:11:57.093543Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation a587e94d-0051-49f0-9fe2-5f29f904ea42 · outbound

This paper cites DeepSeek-V3 Technical Report.

Understanding Fact Recall in Language Models: Why Two-Stage Training Encourages Memorization but Mixed Training Teaches Knowledge DeepSeek-V3 Technical Report

Reference 17

Resolution: unresolved; no resolver link, observed 2026-08-07T15:11:57.187657Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 8cd11863-fa31-430d-9e75-cd0bb50109e5 · outbound

This paper cites Not all lan- guage model features are one-dimensionally linear.

Understanding Fact Recall in Language Models: Why Two-Stage Training Encourages Memorization but Mixed Training Teaches Knowledge Not all lan- guage model features are one-dimensionally linear

Reference 18

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T15:12:02.685879Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation d014f255-0e16-448a-a806-e21eaa130da2 · outbound

This paper cites Dissecting recall of factual associations in auto-regressive language models.

Understanding Fact Recall in Language Models: Why Two-Stage Training Encourages Memorization but Mixed Training Teaches Knowledge Dissecting recall of factual associations in auto-regressive language models

Reference 19

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T15:12:02.670692Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 81677c5e-c886-498b-92d1-b341063cfaa4 · outbound

This paper cites Understanding finetuning for factual knowledge extraction.

Understanding Fact Recall in Language Models: Why Two-Stage Training Encourages Memorization but Mixed Training Teaches Knowledge Understanding finetuning for factual knowledge extraction

Reference 20

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T15:12:02.653178Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 6525eaa8-741a-46dd-9d05-2d331af21aa8 · outbound

This paper cites Reverse training to nurse the reversal curse.

Understanding Fact Recall in Language Models: Why Two-Stage Training Encourages Memorization but Mixed Training Teaches Knowledge Reverse training to nurse the reversal curse

Reference 21

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T15:12:02.636324Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 0c080acd-9d5b-4cbf-9ee5-09248da88793 · outbound

This paper cites Universal neurons in GPT2 language models.Transactions on Machine Learning Research, 2024.

Understanding Fact Recall in Language Models: Why Two-Stage Training Encourages Memorization but Mixed Training Teaches Knowledge Universal neurons in GPT2 language models.Transactions on Machine Learning Research, 2024

Reference 22

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T15:12:02.620553Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation b3a2cf25-4c7d-4583-9365-fe5ad4d26828 · outbound

This paper cites Finding neurons in a haystack: Case studies with sparse probing.

Understanding Fact Recall in Language Models: Why Two-Stage Training Encourages Memorization but Mixed Training Teaches Knowledge Finding neurons in a haystack: Case studies with sparse probing

Reference 23

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T15:12:02.603736Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 8bc5750b-3486-45c9-867a-40dd8d54e304 · outbound

This paper cites Language models represent space and time.

Understanding Fact Recall in Language Models: Why Two-Stage Training Encourages Memorization but Mixed Training Teaches Knowledge Language models represent space and time

Reference 24

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T15:12:02.587439Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 22c0e65d-3d6c-4eb0-816d-d556c3b90c6f · outbound

This paper cites Language models as knowledge bases: On entity rep- resentations, storage capacity, and paraphrased queries.

Understanding Fact Recall in Language Models: Why Two-Stage Training Encourages Memorization but Mixed Training Teaches Knowledge Language models as knowledge bases: On entity rep- resentations, storage capacity, and paraphrased queries

Reference 25

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T15:12:02.571711Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 973a03e0-521d-4407-b949-a848c30e6469 · outbound

This paper cites Monotonic representation of numeric attributes in language models.

Understanding Fact Recall in Language Models: Why Two-Stage Training Encourages Memorization but Mixed Training Teaches Knowledge Monotonic representation of numeric attributes in language models

Reference 26

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T15:12:02.555651Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 7e95b614-8cc0-44ea-97da-1c9198c764b0 · outbound

This paper cites Linearity of relation decoding in transformer language models.

Understanding Fact Recall in Language Models: Why Two-Stage Training Encourages Memorization but Mixed Training Teaches Knowledge Linearity of relation decoding in transformer language models

Reference 27

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T15:12:02.539743Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 20f5776f-811d-4278-b22c-e03aa1035aa0 · outbound

This paper cites Dick, Hidenori Tanaka, Tim Rocktäschel, Edward Grefenstette, and David Krueger.

Understanding Fact Recall in Language Models: Why Two-Stage Training Encourages Memorization but Mixed Training Teaches Knowledge Dick, Hidenori Tanaka, Tim Rocktäschel, Edward Grefenstette, and David Krueger

Reference 28

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T15:12:02.520710Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 3e72995e-9a5b-4270-9341-77e4e3c04a08 · outbound

This paper cites Instruction-tuned language models are better knowledge learners.

Understanding Fact Recall in Language Models: Why Two-Stage Training Encourages Memorization but Mixed Training Teaches Knowledge Instruction-tuned language models are better knowledge learners

Reference 29

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T15:12:02.502964Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 6527b1a8-f8f4-4b71-b4c3-b756d54108a3 · outbound

This paper cites Backward lens: Projecting language model gradients into the vocabulary space.

Understanding Fact Recall in Language Models: Why Two-Stage Training Encourages Memorization but Mixed Training Teaches Knowledge Backward lens: Projecting language model gradients into the vocabulary space

Reference 30

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T15:12:02.486954Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation a4e94535-4f2c-4890-92bc-8106a40b3e26 · outbound

This paper cites The remarkable robustness of LLMs: Stages of inference? In ICML 2024 Workshop on Mechanistic Interpretability, 2024.

Understanding Fact Recall in Language Models: Why Two-Stage Training Encourages Memorization but Mixed Training Teaches Knowledge The remarkable robustness of LLMs: Stages of inference? In ICML 2024 Workshop on Mechanistic Interpretability, 2024

Reference 31

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T15:12:02.470762Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 5697e7b4-3679-4e45-8f6c-e3dda5db059b · outbound

This paper cites Understanding Neural Networks through Representation Erasure.

Understanding Fact Recall in Language Models: Why Two-Stage Training Encourages Memorization but Mixed Training Teaches Knowledge Understanding Neural Networks through Representation Erasure

Reference 32

Resolution: unresolved; no resolver link, observed 2026-08-07T15:11:58.343008Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation f75e0658-dc31-4ad4-b295-7a3d5e5e856f · outbound

This paper cites Relation also knows: Rethinking the recall and editing of factual associations in auto-regressive transformer language models.

Understanding Fact Recall in Language Models: Why Two-Stage Training Encourages Memorization but Mixed Training Teaches Knowledge Relation also knows: Rethinking the recall and editing of factual associations in auto-regressive transformer language models

Reference 33

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T15:12:02.452936Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 7608ad4a-b913-4aa0-8d5e-a3094aef8c41 · outbound

This paper cites The flan collection: Designing data and methods for effective instruction tuning.

Understanding Fact Recall in Language Models: Why Two-Stage Training Encourages Memorization but Mixed Training Teaches Knowledge The flan collection: Designing data and methods for effective instruction tuning

Reference 34

Resolution: unresolved; no resolver link, observed 2026-08-07T15:11:58.484298Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation d0a9076d-61a8-4aeb-82bf-3eb84095e623 · outbound

This paper cites Can neural network memorization be localized? In International Conference on Machine Learning, pages 23536–23557.

Understanding Fact Recall in Language Models: Why Two-Stage Training Encourages Memorization but Mixed Training Teaches Knowledge Can neural network memorization be localized? In International Conference on Machine Learning, pages 23536–23557

Reference 35

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T15:12:02.425608Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 0f22006f-a7a4-4463-b145-45de7dda21d2 · outbound

This paper cites The geometry of truth: Emergent linear structure in large language model representations of true/false datasets.

Understanding Fact Recall in Language Models: Why Two-Stage Training Encourages Memorization but Mixed Training Teaches Knowledge The geometry of truth: Emergent linear structure in large language model representations of true/false datasets

Reference 36

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T15:12:02.408933Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 324e89a1-f10f-4586-9dd0-570cf88c1f6d · outbound

This paper cites Locating and editing factual associations in gpt.

Understanding Fact Recall in Language Models: Why Two-Stage Training Encourages Memorization but Mixed Training Teaches Knowledge Locating and editing factual associations in gpt

Reference 37

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T15:12:02.393825Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 5a7c8702-a860-4522-b987-39873f52e014 · outbound

This paper cites What does the knowledge neuron thesis have to do with knowledge? In International Conference on Learning Representations, 2024.

Understanding Fact Recall in Language Models: Why Two-Stage Training Encourages Memorization but Mixed Training Teaches Knowledge What does the knowledge neuron thesis have to do with knowledge? In International Conference on Learning Representations, 2024

Reference 38

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T15:12:02.377428Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 6678ca99-ffef-4e82-804c-8c5b97384f26 · outbound

This paper cites Interpreting gpt: the logit lens.

Understanding Fact Recall in Language Models: Why Two-Stage Training Encourages Memorization but Mixed Training Teaches Knowledge Interpreting gpt: the logit lens

Reference 39

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T15:12:02.361530Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation b1f5d9e1-5f8e-437e-b61a-8ff6e097dda6 · outbound

This paper cites Competition of mechanisms: Tracing how language models handle facts and coun- terfactuals.

Understanding Fact Recall in Language Models: Why Two-Stage Training Encourages Memorization but Mixed Training Teaches Knowledge Competition of mechanisms: Tracing how language models handle facts and coun- terfactuals

Reference 40

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T15:12:02.345678Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 1fe10073-079a-4996-b57d-7c858aaaf69b · outbound

This paper cites Training language models to follow instructions with human feedback.

Understanding Fact Recall in Language Models: Why Two-Stage Training Encourages Memorization but Mixed Training Teaches Knowledge Training language models to follow instructions with human feedback

Reference 41

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T15:12:02.328998Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 0e1be8a5-bc2b-419d-886e-55c39f5f529b · outbound

This paper cites Task-specific skill localization in fine-tuned language models.

Understanding Fact Recall in Language Models: Why Two-Stage Training Encourages Memorization but Mixed Training Teaches Knowledge Task-specific skill localization in fine-tuned language models

Reference 42

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T15:12:02.312987Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 9870608e-4c31-46cd-87f6-429a3e0f26ef · outbound

This paper cites When do prompting and prefix-tuning work? a theory of capabilities and limitations.

Understanding Fact Recall in Language Models: Why Two-Stage Training Encourages Memorization but Mixed Training Teaches Knowledge When do prompting and prefix-tuning work? a theory of capabilities and limitations

Reference 43

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T15:12:02.295203Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation aabb615d-e817-4643-926d-1a14022edaf1 · outbound

This paper cites Fine-tuning enhances existing mechanisms: A case study on entity tracking.

Understanding Fact Recall in Language Models: Why Two-Stage Training Encourages Memorization but Mixed Training Teaches Knowledge Fine-tuning enhances existing mechanisms: A case study on entity tracking

Reference 44

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T15:12:02.277195Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation e585837d-00c9-4ff6-8a29-67d978c42918 · outbound

This paper cites Neurons in large language models: Dead, n-gram, positional.

Understanding Fact Recall in Language Models: Why Two-Stage Training Encourages Memorization but Mixed Training Teaches Knowledge Neurons in large language models: Dead, n-gram, positional

Reference 45

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T15:12:02.261278Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation bb789b86-5448-4e13-9ad5-83fd90a7f718 · outbound

This paper cites Finding skill neurons in pre-trained transformer-based language models.

Understanding Fact Recall in Language Models: Why Two-Stage Training Encourages Memorization but Mixed Training Teaches Knowledge Finding skill neurons in pre-trained transformer-based language models

Reference 46

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T15:12:02.225231Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 1ee714e2-6c90-4c38-9405-4de4a732c500 · outbound

This paper cites Super-NaturalInstructions: Generalization via declarative instructions on 1600+ NLP tasks.

Understanding Fact Recall in Language Models: Why Two-Stage Training Encourages Memorization but Mixed Training Teaches Knowledge Super-NaturalInstructions: Generalization via declarative instructions on 1600+ NLP tasks

Reference 47

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T15:12:02.088541Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation af953b96-c6e4-4fb9-a0a5-38bd1f900ba8 · outbound

This paper cites Qwen2.5 Technical Report.

Understanding Fact Recall in Language Models: Why Two-Stage Training Encourages Memorization but Mixed Training Teaches Knowledge Qwen2.5 Technical Report

Reference 48

Resolution: unresolved; no resolver link, observed 2026-08-07T15:12:00.106658Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 90f0e5b7-c7bd-460d-8905-ab3020ec4f4a · outbound

This paper cites Knowledge circuits in pretrained transformers.

Understanding Fact Recall in Language Models: Why Two-Stage Training Encourages Memorization but Mixed Training Teaches Knowledge Knowledge circuits in pretrained transformers

Reference 49

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T15:12:01.921457Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation e6d827a9-bbaa-47a2-beb1-ddb678f13b58 · outbound

This paper cites Towards best practices of activation patching in language models: Metrics and methods.

Understanding Fact Recall in Language Models: Why Two-Stage Training Encourages Memorization but Mixed Training Teaches Knowledge Towards best practices of activation patching in language models: Metrics and methods

Reference 50

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T15:12:01.782256Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation c32fb028-0170-4ef5-94c5-4b4add4ab808 · outbound

This paper cites How do large language models handle multilingualism? In Advances in Neural Information Processing Systems, 2024.

Understanding Fact Recall in Language Models: Why Two-Stage Training Encourages Memorization but Mixed Training Teaches Knowledge How do large language models handle multilingualism? In Advances in Neural Information Processing Systems, 2024

Reference 51

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T15:12:01.686449Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 5612ee6c-37a7-423d-a832-82276e6275da · outbound

This paper cites (b) Overlapping Ratio vs.

Understanding Fact Recall in Language Models: Why Two-Stage Training Encourages Memorization but Mixed Training Teaches Knowledge (b) Overlapping Ratio vs

Reference 52

Resolution: malformed identifier; raw_fallback, observed 2026-08-07T15:12:00.838220Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation d2c135a1-8043-40b7-baa0-4589f0f8e668 · outbound

This paper cites He benefited from the world-class education and research facilities at Andrew Jackson University.

Understanding Fact Recall in Language Models: Why Two-Stage Training Encourages Memorization but Mixed Training Teaches Knowledge He benefited from the world-class education and research facilities at Andrew Jackson University

Reference 54

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T15:12:01.182317Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation f8f2214e-c568-46ed-bedc-a491e28c43e7 · outbound

This paper cites He benefited from the world-class education and research facilities at Andrew Jackson University.

Reference 1982

Resolution: verified fuzzy; raw_fallback, observed 2026-08-07T15:12:01.362715Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Pith citing papers

Observation aa3b18c5-23ec-47e6-9973-0621f72de720 · inbound

Reverse Convolution and Its Applications to Image Restoration cites this paper.

Reverse Convolution and Its Applications to Image Restoration Understanding Fact Recall in Language Models: Why Two-Stage Training Encourages Memorization but Mixed Training Teaches Knowledge

Reference 6

Resolution: unresolved; no resolver link, observed 2026-08-05T20:53:40.719452Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 05210f89-4f2d-46d8-bc3a-8c44449424cb · inbound

Deep sequence models tend to memorize geometrically; it is unclear why cites this paper.

Deep sequence models tend to memorize geometrically; it is unclear why Understanding Fact Recall in Language Models: Why Two-Stage Training Encourages Memorization but Mixed Training Teaches Knowledge

Reference 202

Resolution: verified exact; arxiv_id, observed 2026-05-29T02:04:53.815446Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.

Observation 3034ac67-e9b9-42a3-b4a9-feed42715713 · inbound

LMs as Task-Specific Knowledge Bases: An Interpretability Analysis cites this paper.

LMs as Task-Specific Knowledge Bases: An Interpretability Analysis Understanding Fact Recall in Language Models: Why Two-Stage Training Encourages Memorization but Mixed Training Teaches Knowledge

Reference 71

Resolution: metadata mismatch; local_arxiv, observed 2026-07-04T14:19:53.914287Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-10T06:31:04.303077+00:00.