Paper Citation Record · LEDGER

Scalpel vs. Hammer: GRPO Amplifies Existing Capabilities, SFT Replaces Them

As of 7 August 2026, this Paper Citation Record lists 48 of 48 outbound references and 5 inbound Pith citation observations for arXiv:2507.10616.

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

pith.paper-citation-record.v1
2507.10616 v2

Coverage vector

measured 48 of 48 reference resolution

Typed states for the displayed outbound observations.

Source: paper_references, paper_reference_links, observed 2026-08-06T17:52:01.734792Z

measured 53 of 53 standing notices

One-hop event checks from named stored sources.

Source: scholarly_work_events, retraction_status_cache, observed 2026-08-07T06:34:17.273281+00:00

measured 5 of 5 inbound itemization

Pith citing papers itemized under the disclosed page cap.

Source: paper_references, paper_reference_links, observed 2026-08-04T07:21:36.979706Z

measured 0 of 1 external citation measurements

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

Source: arxiv_reference, observed 2026-05-18T13:31:24.957001Z

Reference resolution

48 of 48 outbound references displayed

verified exact0
verified fuzzy9
unresolved39
parse uncertain0
malformed identifier0
metadata mismatch0

External citation measurements

No source-named external measurement is stored.

Outbound references

Observation 15f1e1a1-4a69-4f00-97d0-79b3a640d466 · outbound

This paper cites Mechanistic Interpretability for AI Safety -- A Review.

Scalpel vs. Hammer: GRPO Amplifies Existing Capabilities, SFT Replaces Them Mechanistic Interpretability for AI Safety -- A Review

Reference 3

Resolution: unresolved; no resolver link, observed 2026-08-06T17:52:01.541750Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 2275602e-b19f-449a-a49b-510d45f03549 · outbound

This paper cites Iteration Head: A Mechanistic Study of Chain-of-Thought.

Scalpel vs. Hammer: GRPO Amplifies Existing Capabilities, SFT Replaces Them Iteration Head: A Mechanistic Study of Chain-of-Thought

Reference 6

Resolution: unresolved; no resolver link, observed 2026-08-06T17:52:01.557005Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 57e9a16c-e03b-40da-99b0-e566e1ccce71 · outbound

This paper cites Rethinking Reflection in Pre-Training.

Scalpel vs. Hammer: GRPO Amplifies Existing Capabilities, SFT Replaces Them Rethinking Reflection in Pre-Training

Reference 8

Resolution: unresolved; no resolver link, observed 2026-08-06T17:52:01.565519Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 58395d0a-9847-4558-8b9d-471c84e687ec · outbound

This paper cites Cognitive Behaviors that Enable Self-Improving Reasoners, or, Four Habits of Highly Effective STaRs.

Scalpel vs. Hammer: GRPO Amplifies Existing Capabilities, SFT Replaces Them Cognitive Behaviors that Enable Self-Improving Reasoners, or, Four Habits of Highly Effective STaRs

Reference 9

Resolution: unresolved; no resolver link, observed 2026-08-06T17:52:01.569741Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 031badbe-1f1f-4efe-9197-02f111fce574 · outbound

This paper cites Geva, M., Schuster, R., Berant, J., and Levy, O.

Scalpel vs. Hammer: GRPO Amplifies Existing Capabilities, SFT Replaces Them Geva, M., Schuster, R., Berant, J., and Levy, O

Reference 10

Resolution: verified fuzzy; raw_fallback, observed 2026-08-06T17:52:02.607100Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation 8d97fc5f-7d92-4f99-b5ac-d69f97d2b1b4 · outbound

This paper cites Transformer Feed-Forward Layers Are Key-Value Memories.

Scalpel vs. Hammer: GRPO Amplifies Existing Capabilities, SFT Replaces Them Transformer Feed-Forward Layers Are Key-Value Memories

Reference 11

Resolution: unresolved; no resolver link, observed 2026-08-06T17:52:01.578986Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 3daef553-5693-42c1-b2be-dded08610a88 · outbound

This paper cites Gugger, S., Debut, L., Wolf, T., Schmid, P., Mueller, Z., Mangrulkar, S., Sun, M., and Bossan, B.

Scalpel vs. Hammer: GRPO Amplifies Existing Capabilities, SFT Replaces Them Gugger, S., Debut, L., Wolf, T., Schmid, P., Mueller, Z., Mangrulkar, S., Sun, M., and Bossan, B

Reference 12

Resolution: verified fuzzy; raw_fallback, observed 2026-08-06T17:52:02.594005Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation aefa57ea-aeb5-415b-8397-01a463fa3bba · outbound

This paper cites Learning to Reason for Long-Form Story Generation.

Scalpel vs. Hammer: GRPO Amplifies Existing Capabilities, SFT Replaces Them Learning to Reason for Long-Form Story Generation

Reference 13

Resolution: unresolved; no resolver link, observed 2026-08-06T17:52:01.587596Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 08731407-9b6c-4b62-a811-98143761663b · outbound

This paper cites Does Localization Inform Editing? Surprising Differences in Causality-Based Localization vs. Knowledge Editing in Language Models.

Scalpel vs. Hammer: GRPO Amplifies Existing Capabilities, SFT Replaces Them Does Localization Inform Editing? Surprising Differences in Causality-Based Localization vs. Knowledge Editing in Language Models

Reference 14

Resolution: unresolved; no resolver link, observed 2026-08-06T17:52:01.591918Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 72d2d0e0-4ccc-4476-aece-eaf3603403bf · outbound

This paper cites Measuring Massive Multitask Language Understanding.

Scalpel vs. Hammer: GRPO Amplifies Existing Capabilities, SFT Replaces Them Measuring Massive Multitask Language Understanding

Reference 15

Resolution: unresolved; no resolver link, observed 2026-08-06T17:52:01.596734Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation bca38365-6914-41bd-8d5e-6f8894086a43 · outbound

This paper cites Huang, Z., Cheng, T., Qiu, Z., Wang, Z., Xu, Y ., Ponti, E.

Scalpel vs. Hammer: GRPO Amplifies Existing Capabilities, SFT Replaces Them Huang, Z., Cheng, T., Qiu, Z., Wang, Z., Xu, Y ., Ponti, E

Reference 16

Resolution: unresolved; no resolver link, observed 2026-08-06T17:52:01.600906Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation ea22cde7-d76b-4c3d-afb5-d19f799e5c6d · outbound

This paper cites Blending Supervised and Reinforcement Fine-Tuning with Prefix Sampling.

Scalpel vs. Hammer: GRPO Amplifies Existing Capabilities, SFT Replaces Them Blending Supervised and Reinforcement Fine-Tuning with Prefix Sampling

Reference 17

Resolution: unresolved; no resolver link, observed 2026-08-06T17:52:01.604834Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation a0f232f5-438e-44be-92f3-47389c9131f5 · outbound

This paper cites Tulu 3: Pushing Frontiers in Open Language Model Post-Training.

Scalpel vs. Hammer: GRPO Amplifies Existing Capabilities, SFT Replaces Them Tulu 3: Pushing Frontiers in Open Language Model Post-Training

Reference 18

Resolution: unresolved; no resolver link, observed 2026-08-06T17:52:01.609100Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 3b0fa0d9-b09a-4ac6-8706-3bea129418de · outbound

This paper cites Let's Verify Step by Step.

Scalpel vs. Hammer: GRPO Amplifies Existing Capabilities, SFT Replaces Them Let's Verify Step by Step

Reference 19

Resolution: unresolved; no resolver link, observed 2026-08-06T17:52:01.613430Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation fed83c84-368a-4e0a-9722-2b33a67bb271 · outbound

This paper cites The Llama 3 Herd of Models.

Scalpel vs. Hammer: GRPO Amplifies Existing Capabilities, SFT Replaces Them The Llama 3 Herd of Models

Reference 20

Resolution: unresolved; no resolver link, observed 2026-08-06T17:52:01.617778Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 330fffd6-db9c-4c8a-b5bc-8ed779e39786 · outbound

This paper cites Locating and Editing Factual Associations in GPT.

Scalpel vs. Hammer: GRPO Amplifies Existing Capabilities, SFT Replaces Them Locating and Editing Factual Associations in GPT

Reference 21

Resolution: unresolved; no resolver link, observed 2026-08-06T17:52:01.622097Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 7d84c587-901a-4af9-8ab1-912bad785981 · outbound

This paper cites org/abs/2505.11711.

Scalpel vs. Hammer: GRPO Amplifies Existing Capabilities, SFT Replaces Them org/abs/2505.11711

Reference 22

Resolution: unresolved; no resolver link, observed 2026-08-06T17:52:01.626268Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 26ffe8c6-2434-40bd-8536-f992215c10bc · outbound

This paper cites OpenAI o1 System Card.

Scalpel vs. Hammer: GRPO Amplifies Existing Capabilities, SFT Replaces Them OpenAI o1 System Card

Reference 24

Resolution: unresolved; no resolver link, observed 2026-08-06T17:52:01.633869Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 9de76864-28a6-4bdd-89f3-c44fc9d00bb9 · outbound

This paper cites Prakash, N., Shaham, T.

Scalpel vs. Hammer: GRPO Amplifies Existing Capabilities, SFT Replaces Them Prakash, N., Shaham, T

Reference 25

Resolution: verified fuzzy; raw_fallback, observed 2026-08-06T17:52:02.566044Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation 7af43f6a-da75-4816-85b6-e438c219a61a · outbound

This paper cites Fine-Tuning Enhances Existing Mechanisms: A Case Study on Entity Tracking.

Scalpel vs. Hammer: GRPO Amplifies Existing Capabilities, SFT Replaces Them Fine-Tuning Enhances Existing Mechanisms: A Case Study on Entity Tracking

Reference 26

Resolution: unresolved; no resolver link, observed 2026-08-06T17:52:01.642501Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 1ada028b-d9c6-46a9-b952-f09e2fd262f5 · outbound

This paper cites ZeRO: Memory Optimizations Toward Training Trillion Parameter Models.

Scalpel vs. Hammer: GRPO Amplifies Existing Capabilities, SFT Replaces Them ZeRO: Memory Optimizations Toward Training Trillion Parameter Models

Reference 27

Resolution: unresolved; no resolver link, observed 2026-08-06T17:52:01.646838Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 38372a48-07c4-492c-8a09-61d70d6f4f4f · outbound

This paper cites GPQA: A Graduate-Level Google-Proof Q&A Benchmark.

Scalpel vs. Hammer: GRPO Amplifies Existing Capabilities, SFT Replaces Them GPQA: A Graduate-Level Google-Proof Q&A Benchmark

Reference 28

Resolution: unresolved; no resolver link, observed 2026-08-06T17:52:01.651019Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 534a4077-01a5-4494-a7a5-bf58eb1145a6 · outbound

This paper cites DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models.

Scalpel vs. Hammer: GRPO Amplifies Existing Capabilities, SFT Replaces Them DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models

Reference 29

Resolution: unresolved; no resolver link, observed 2026-08-06T17:52:01.655006Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 401043c3-2749-44ab-a332-cf595f14b532 · outbound

This paper cites Amuro and Char: Analyzing the Relationship between Pre-Training and Fine-Tuning of Large Language Models.

Scalpel vs. Hammer: GRPO Amplifies Existing Capabilities, SFT Replaces Them Amuro and Char: Analyzing the Relationship between Pre-Training and Fine-Tuning of Large Language Models

Reference 30

Resolution: unresolved; no resolver link, observed 2026-08-06T17:52:01.659109Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation dc56e706-31a6-4160-9c14-1dbfeae88be1 · outbound

This paper cites Optimizing Language Models for Inference Time Objectives using Reinforcement Learning.

Scalpel vs. Hammer: GRPO Amplifies Existing Capabilities, SFT Replaces Them Optimizing Language Models for Inference Time Objectives using Reinforcement Learning

Reference 31

Resolution: unresolved; no resolver link, observed 2026-08-06T17:52:01.663454Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation fa093905-f19a-41f1-8a2d-452069abcbde · outbound

This paper cites 2 OLMo 2 Furious.

Scalpel vs. Hammer: GRPO Amplifies Existing Capabilities, SFT Replaces Them 2 OLMo 2 Furious

Reference 32

Resolution: unresolved; no resolver link, observed 2026-08-06T17:52:01.667668Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation d0447302-44f7-47e5-92e4-bf6fb966b106 · outbound

This paper cites LLM Circuit Analyses Are Consistent Across Training and Scale.

Scalpel vs. Hammer: GRPO Amplifies Existing Capabilities, SFT Replaces Them LLM Circuit Analyses Are Consistent Across Training and Scale

Reference 33

Resolution: unresolved; no resolver link, observed 2026-08-06T17:52:01.671958Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 0d1d4c40-723c-48b0-92a5-ca7436e0cd66 · outbound

This paper cites Interpretability in the Wild: a Circuit for Indirect Object Identification in GPT-2 small.

Scalpel vs. Hammer: GRPO Amplifies Existing Capabilities, SFT Replaces Them Interpretability in the Wild: a Circuit for Indirect Object Identification in GPT-2 small

Reference 34

Resolution: unresolved; no resolver link, observed 2026-08-06T17:52:01.676363Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 600ea502-3724-4313-bc88-ff89b642cfd6 · outbound

This paper cites MMLU-Pro: A More Robust and Challenging Multi-Task Language Understanding Benchmark.

Scalpel vs. Hammer: GRPO Amplifies Existing Capabilities, SFT Replaces Them MMLU-Pro: A More Robust and Challenging Multi-Task Language Understanding Benchmark

Reference 35

Resolution: unresolved; no resolver link, observed 2026-08-06T17:52:01.680504Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 57ba9d02-d990-4ef2-b027-ec836295e367 · outbound

This paper cites Chain-of-Thought Prompting Elicits Reasoning in Large Language Models.

Scalpel vs. Hammer: GRPO Amplifies Existing Capabilities, SFT Replaces Them Chain-of-Thought Prompting Elicits Reasoning in Large Language Models

Reference 36

Resolution: unresolved; no resolver link, observed 2026-08-06T17:52:01.684409Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 91d2abdc-f7df-4bb8-b0f2-144860d76688 · outbound

This paper cites Learning to Reason under Off-Policy Guidance.

Scalpel vs. Hammer: GRPO Amplifies Existing Capabilities, SFT Replaces Them Learning to Reason under Off-Policy Guidance

Reference 37

Resolution: unresolved; no resolver link, observed 2026-08-06T17:52:01.688475Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation aa625e0e-132a-47e6-85d8-bb3c5d5d49a8 · outbound

This paper cites DAPO: An Open-Source LLM Reinforcement Learning System at Scale.

Scalpel vs. Hammer: GRPO Amplifies Existing Capabilities, SFT Replaces Them DAPO: An Open-Source LLM Reinforcement Learning System at Scale

Reference 38

Resolution: unresolved; no resolver link, observed 2026-08-06T17:52:01.692858Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 72bd5291-34fa-4321-bfc2-ecbb7b02bac0 · outbound

This paper cites Does Reinforcement Learning Really Incentivize Reasoning Capacity in LLMs Beyond the Base Model?.

Scalpel vs. Hammer: GRPO Amplifies Existing Capabilities, SFT Replaces Them Does Reinforcement Learning Really Incentivize Reasoning Capacity in LLMs Beyond the Base Model?

Reference 39

Resolution: unresolved; no resolver link, observed 2026-08-06T17:52:01.696842Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 6bc7292e-75cf-4773-bb90-8c07c170c430 · outbound

This paper cites Echo Chamber: RL Post-training Amplifies Behaviors Learned in Pretraining.

Scalpel vs. Hammer: GRPO Amplifies Existing Capabilities, SFT Replaces Them Echo Chamber: RL Post-training Amplifies Behaviors Learned in Pretraining

Reference 40

Resolution: unresolved; no resolver link, observed 2026-08-06T17:52:01.701241Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 92339cb4-1015-41e6-a58c-3758845cbf7d · outbound

This paper cites Instruction-Following Evaluation for Large Language Models.

Scalpel vs. Hammer: GRPO Amplifies Existing Capabilities, SFT Replaces Them Instruction-Following Evaluation for Large Language Models

Reference 41

Resolution: unresolved; no resolver link, observed 2026-08-06T17:52:01.705308Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 46966d1d-5e86-445c-b6d4-ad28d16f0c43 · outbound

This paper cites an unresolved cited work.

Scalpel vs. Hammer: GRPO Amplifies Existing Capabilities, SFT Replaces Them Unresolved cited work

Reference 42

Resolution: unresolved; raw_fallback, observed 2026-08-06T17:52:02.552092Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation b0f3f168-9145-4c8c-af2c-398a45c99ff8 · outbound

This paper cites In the former, the model continuously struggles on many of the questions, while in the latter, the model gradually exhibits much stronger performance.

Scalpel vs. Hammer: GRPO Amplifies Existing Capabilities, SFT Replaces Them In the former, the model continuously struggles on many of the questions, while in the latter, the model gradually exhibits much stronger performance

Reference 43

Resolution: verified fuzzy; raw_fallback, observed 2026-08-06T17:52:02.536594Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation d51d9639-f7cc-4a21-a8bc-ff3f3d40ea6b · outbound

This paper cites Note that Shao et al.

Scalpel vs. Hammer: GRPO Amplifies Existing Capabilities, SFT Replaces Them Note that Shao et al

Reference 44

Resolution: verified fuzzy; raw_fallback, observed 2026-08-06T17:52:02.520350Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation 42647dd1-47e2-47b6-aed2-9b2160de499a · outbound

This paper cites an unresolved cited work.

Scalpel vs. Hammer: GRPO Amplifies Existing Capabilities, SFT Replaces Them Unresolved cited work

Reference 45

Resolution: unresolved; raw_fallback, observed 2026-08-06T17:52:02.506604Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation 8f6f5c16-524c-4492-907d-020485f8da59 · outbound

This paper cites system prompt.

Scalpel vs. Hammer: GRPO Amplifies Existing Capabilities, SFT Replaces Them system prompt

Reference 46

Resolution: verified fuzzy; raw_fallback, observed 2026-08-06T17:52:02.493208Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation 388a5d13-e64d-47db-9b9b-bdc517c4fb2a · outbound

This paper cites an unresolved cited work.

Scalpel vs. Hammer: GRPO Amplifies Existing Capabilities, SFT Replaces Them Unresolved cited work

Reference 47

Resolution: unresolved; raw_fallback, observed 2026-08-06T17:52:02.479269Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation 5c15a31e-bcc7-4d3d-a602-cba444aba355 · outbound

This paper cites We use the same exact hyperparameters and dataset as Meng et al.

Scalpel vs. Hammer: GRPO Amplifies Existing Capabilities, SFT Replaces Them We use the same exact hyperparameters and dataset as Meng et al

Reference 48

Resolution: verified fuzzy; raw_fallback, observed 2026-08-06T17:52:02.465531Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation 07ec9d49-a608-4571-bad1-0d2dff4684cd · outbound

This paper cites Language Models are Few-Shot Learners.

Scalpel vs. Hammer: GRPO Amplifies Existing Capabilities, SFT Replaces Them Language Models are Few-Shot Learners

Reference 2020

Resolution: unresolved; no resolver link, observed 2026-08-06T17:52:01.550647Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation f1a22faf-31a5-4f70-a710-307f2abda1f2 · outbound

This paper cites EssentialAI, :, Shah, D.

Scalpel vs. Hammer: GRPO Amplifies Existing Capabilities, SFT Replaces Them EssentialAI, :, Shah, D

Reference 2021

Resolution: verified fuzzy; raw_fallback, observed 2026-08-06T17:52:02.620016Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation e4b1d23f-2d53-41e8-bed1-3ffa96005d37 · outbound

This paper cites an unresolved cited work.

Scalpel vs. Hammer: GRPO Amplifies Existing Capabilities, SFT Replaces Them Unresolved cited work

Reference 2022

Resolution: unresolved; raw_fallback, observed 2026-08-06T17:52:02.579915Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation dd1ae8f0-1cf2-4cb6-bdfc-e0bc230b9495 · outbound

This paper cites Pythia: A Suite for Analyzing Large Language Models Across Training and Scaling.

Scalpel vs. Hammer: GRPO Amplifies Existing Capabilities, SFT Replaces Them Pythia: A Suite for Analyzing Large Language Models Across Training and Scaling

Reference 2023

Resolution: unresolved; no resolver link, observed 2026-08-06T17:52:01.546209Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation d746061d-7f36-4a70-ad54-5efb82cc6032 · outbound

This paper cites SmolLM2: When Smol Goes Big -- Data-Centric Training of a Small Language Model.

Scalpel vs. Hammer: GRPO Amplifies Existing Capabilities, SFT Replaces Them SmolLM2: When Smol Goes Big -- Data-Centric Training of a Small Language Model

Reference 2024

Resolution: unresolved; no resolver link, observed 2026-08-06T17:52:01.531978Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation b25e3c51-48d8-48ea-8c84-54e04df4cc25 · outbound

This paper cites Bereska, L.

Scalpel vs. Hammer: GRPO Amplifies Existing Capabilities, SFT Replaces Them Bereska, L

Reference 2025

Resolution: verified fuzzy; raw_fallback, observed 2026-08-06T17:52:02.633127Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Pith citing papers

Observation da915c00-d435-4c7e-9eec-e2466bc44ada · inbound

Thinking Sparks!: Emergent Attention Heads in Reasoning Models During Post Training cites this paper.

Thinking Sparks!: Emergent Attention Heads in Reasoning Models During Post Training Scalpel vs. Hammer: GRPO Amplifies Existing Capabilities, SFT Replaces Them

Reference 31

Resolution: verified exact; arxiv_id, observed 2026-05-18T13:31:24.958893Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation bbe44da7-e415-4b51-972f-25ad4f82d755 · inbound

Value Drifts: Tracing Value Alignment During LLM Post-Training cites this paper.

Value Drifts: Tracing Value Alignment During LLM Post-Training Scalpel vs. Hammer: GRPO Amplifies Existing Capabilities, SFT Replaces Them

Reference 57

Resolution: unresolved; no resolver link, observed 2026-08-04T07:21:36.979706Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 6a028c1d-a1f0-43e3-8bb6-79290508172e · inbound

EMO-R3: Reflective Reinforcement Learning for Emotional Reasoning in Multimodal Large Language Models cites this paper.

EMO-R3: Reflective Reinforcement Learning for Emotional Reasoning in Multimodal Large Language Models Scalpel vs. Hammer: GRPO Amplifies Existing Capabilities, SFT Replaces Them

Reference 40

Resolution: unresolved; no resolver link, observed 2026-08-02T20:14:04.093885Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 669a0cc0-8587-4c91-948b-cb9ebc6c8502 · inbound

TOPCELL: Topology Optimization of Standard Cell via LLMs cites this paper.

TOPCELL: Topology Optimization of Standard Cell via LLMs Scalpel vs. Hammer: GRPO Amplifies Existing Capabilities, SFT Replaces Them

Reference 27

Resolution: metadata mismatch; arxiv_id, observed 2026-05-10T14:15:29.308434Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.

Observation e8252faa-6deb-4380-bd1e-95bd3d5eda90 · inbound

Decouple before Integration: Test-time Synthesis of SFT and RLVR Task Vectors cites this paper.

Decouple before Integration: Test-time Synthesis of SFT and RLVR Task Vectors Scalpel vs. Hammer: GRPO Amplifies Existing Capabilities, SFT Replaces Them

Reference 35

Resolution: metadata mismatch; arxiv_id, observed 2026-05-11T15:51:44.090681Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-07T06:34:17.273281+00:00.