# Rethinking Data Selection for Supervised Fine-Tuning

arXiv: https://arxiv.org/abs/2402.06094 · HTML: https://mlsanity.com/paper/2402.06094/

## Key facts

- Cited by 35 papers in the map (data through 14 Oct 2026).
- Placed in Reasoning benchmarks and evaluation, one of 17 directions in LLM reasoning.
- Builds on 1 of its 74 references that are in the map.
- First posted on arXiv 8 Feb 2024.

## Placed in

- LLM reasoning / [Reasoning benchmarks and evaluation](https://mlsanity.com/lr/reasoning-benchmarks-and-evaluation/)

## Builds on (most cited first, 1 of 1)

- [AlpaGasus: Training A Better Alpaca with Fewer Data](https://mlsanity.com/paper/2307.08701/) (2023)

## Built on by (most cited first, 30 of 35)

- [LLaVA-CoT: Let Vision Language Models Reason Step-by-Step](https://mlsanity.com/paper/2411.10440/) (2024)
- [A Survey of Self-Evolving Agents: What, When, How, and Where to Evolve on the Path to Artificial Super Intelligence](https://mlsanity.com/paper/2507.21046/) (2025)
- [Intern-S1: A Scientific Multimodal Foundation Model](https://mlsanity.com/paper/2508.15763/) (2025)
- [Stepwise Perplexity-Guided Refinement for Efficient Chain-of-Thought Reasoning in Large Language Models](https://mlsanity.com/paper/2502.13260/) (2025)
- [Entropy Law: The Story Behind Data Compression and LLM Performance](https://mlsanity.com/paper/2407.06645/) (2024)
- [Learning or Self-aligning? Rethinking Instruction Fine-tuning](https://mlsanity.com/paper/2402.18243/) (2024)
- [Safety-Aware Fine-Tuning of Large Language Models](https://mlsanity.com/paper/2410.10014/) (2024)
- [Rethinking Data Selection at Scale: Random Selection is Almost All You Need](https://mlsanity.com/paper/2410.09335/) (2024)
- [The Open Source Advantage in Large Language Models (LLMs)](https://mlsanity.com/paper/2412.12004/) (2024)
- [Condor: Enhance LLM Alignment with Knowledge-Driven Data Synthesis and Refinement](https://mlsanity.com/paper/2501.12273/) (2025)
- [Chiron-o1: Igniting Multimodal Large Language Models towards Generalizable Medical Reasoning via Mentor-Intern Collaborative Search](https://mlsanity.com/paper/2506.16962/) (2025)
- [LLM Roleplay: Simulating Human-Chatbot Interaction](https://mlsanity.com/paper/2407.03974/) (2024)
- [Beyond the Surface: Enhancing LLM-as-a-Judge Alignment with Human via Internal Representations](https://mlsanity.com/paper/2508.03550/) (2025)
- [Chasing Random: Instruction Selection Strategies Fail to Generalize](https://mlsanity.com/paper/2410.15225/) (2024)
- [RePIC: Reinforced Post-Training for Personalizing Multi-Modal Language Models](https://mlsanity.com/paper/2506.18369/) (2025)
- [IDEAL: Data Equilibrium Adaptation for Multi-Capability Language Model Alignment](https://mlsanity.com/paper/2505.12762/) (2025)
- [DeepSound-V1: Start to Think Step-by-Step in the Audio Generation from Videos](https://mlsanity.com/paper/2503.22208/) (2025)
- [Data Selection for Multi-turn Dialogue Instruction Tuning](https://mlsanity.com/paper/2604.07892/) (2026)
- [MergeIT: From Selection to Merging for Efficient Instruction Tuning](https://mlsanity.com/paper/2503.00034/) (2025)
- [TAG-INSTRUCT: Controlled Instruction Complexity Enhancement through Structure-based Augmentation](https://mlsanity.com/paper/2505.18557/) (2025)
- [CircuitSeer: Mining High-Quality Data by Probing Mathematical Reasoning Circuits in LLMs](https://mlsanity.com/paper/2510.18470/) (2025)
- [CLLoRA: An Approach to Measure the Effects of the Context Length for LLM Fine-Tuning](https://mlsanity.com/paper/2502.18910/) (2025)
- [Exploring Solution Divergence and Its Effect on Large Language Model Problem Solving](https://mlsanity.com/paper/2509.22480/) (2025)
- [Computational Budget Should Be Considered in Data Selection](https://mlsanity.com/paper/2510.16806/) (2025)
- [Mistral-C2F: Coarse to Fine Actor for Analytical and Reasoning Enhancement in RLHF and Effective-Merged LLMs](https://mlsanity.com/paper/2406.08657/) (2024)
- [Boosting LLM via Learning from Data Iteratively and Selectively](https://mlsanity.com/paper/2412.17365/) (2024)
- [Beyond QA Pairs: Assessing Parameter-Efficient Fine-Tuning for Fact Embedding in LLMs](https://mlsanity.com/paper/2503.01131/) (2025)
- [V-VAE: A Variational Auto Encoding Framework Towards Fine-Grained Control over Human-Like Chat](https://mlsanity.com/paper/2506.01524/) (2025)
- [bi-GRPO: Bidirectional Optimization for Jailbreak Backdoor Injection on LLMs](https://mlsanity.com/paper/2509.19775/) (2025)
- [Short Chains, Deep Thoughts: Balancing Reasoning Efficiency and Intra-Segment Capability via Split-Merge Optimization](https://mlsanity.com/paper/2602.03141/) (2026)

## Abstract (arXiv)

Although supervised finetuning (SFT) has emerged as an essential technique to align large language models with humans, it is considered superficial, with style learning being its nature. At the same time, recent works indicate the importance of data selection for SFT, showing that finetuning with high-quality and diverse subsets of the original dataset leads to superior downstream performance. In this work, we rethink the intuition behind data selection for SFT. Considering SFT is superficial, we propose that essential demonstrations for SFT should focus on reflecting human-like interactions instead of data quality or diversity. However, it is not straightforward to directly assess to what extent a demonstration reflects human styles. Towards an initial attempt in this direction, we find selecting instances with long responses is surprisingly more effective for SFT than utilizing full datasets or instances selected based on quality and diversity. We hypothesize that such a simple heuristic implicitly mimics a crucial aspect of human-style conversation: detailed responses are usually more helpful.