# Recursive Training Loops in LLMs: How training data properties modulate distribution shift in generated data?

arXiv: https://arxiv.org/abs/2504.03814 · HTML: https://mlsanity.com/paper/2504.03814/

## Key facts

- Cited by 2 papers in the map (data through 14 Oct 2026).
- Builds on 29 of its 52 references that are in the map.
- First posted on arXiv 4 Apr 2025.

## Builds on (most cited first, 29 of 29)

- [The Llama 3 Herd of Models](https://mlsanity.com/paper/2407.21783/) (2024)
- [LoRA: Low-Rank Adaptation of Large Language Models](https://mlsanity.com/paper/2106.09685/) (2021)
- [Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks](https://mlsanity.com/paper/1908.10084/) (2019)
- [A Survey on Large Language Models for Code Generation](https://mlsanity.com/paper/2406.00515/) (2024)
- [Texygen: A Benchmarking Platform for Text Generation Models](https://mlsanity.com/paper/1802.01886/) (2018)
- [Is ChatGPT a Good NLG Evaluator? A Preliminary Study](https://mlsanity.com/paper/2303.04048/) (2023)
- [The Curse of Recursion: Training on Generated Data Makes Models Forget](https://mlsanity.com/paper/2305.17493/) (2023)
- [Open-Ended Learning Leads to Generally Capable Agents](https://mlsanity.com/paper/2107.12808/) (2021)
- [Turning Up the Heat: Min-p Sampling for Creative and Coherent LLM Outputs](https://mlsanity.com/paper/2407.01082/) (2024)
- [A Tale of Tails: Model Collapse as a Change of Scaling Laws](https://mlsanity.com/paper/2402.07043/) (2024)
- [A Note on Shumailov et al. (2024): `AI Models Collapse When Trained on Recursively Generated Data'](https://mlsanity.com/paper/2410.12954/) (2024)
- [Jasper and Stella: distillation of SOTA embedding models](https://mlsanity.com/paper/2412.19048/) (2024)
- [On the Stability of Iterative Retraining of Generative Models on their own Data](https://mlsanity.com/paper/2310.00429/) (2023)
- [The Curious Decline of Linguistic Diversity: Training Language Models on Synthetic Text](https://mlsanity.com/paper/2311.09807/) (2023)
- [The Rise of Social Bots](https://mlsanity.com/paper/1407.5225/) (2014)
- [Exploring the Use of Large Language Models for Reference-Free Text Quality Evaluation: An Empirical Study](https://mlsanity.com/paper/2304.00723/) (2023)
- [Model Collapse Demystified: The Case of Regression](https://mlsanity.com/paper/2402.07712/) (2024)
- [Kickstarting Deep Reinforcement Learning](https://mlsanity.com/paper/1803.03835/) (2018)
- [Large Language Models Suffer From Their Own Output: An Analysis of the Self-Consuming Training Loop](https://mlsanity.com/paper/2311.16822/) (2023)
- [OpenMEVA: A Benchmark for Evaluating Open-ended Story Generation Metrics](https://mlsanity.com/paper/2105.08920/) (2021)
- [Will Large-scale Generative Models Corrupt Future Datasets?](https://mlsanity.com/paper/2211.08095/) (2022)
- [Towards Understanding the Interplay of Generative Artificial Intelligence and the Internet](https://mlsanity.com/paper/2306.06130/) (2023)
- [Collapse or Thrive? Perils and Promises of Synthetic Data in a Self-Generating World](https://mlsanity.com/paper/2410.16713/) (2024)
- [Combining Generative Artificial Intelligence (AI) and the Internet: Heading towards Evolution or Degradation?](https://mlsanity.com/paper/2303.01255/) (2023)
- [Position: Model Collapse Does Not Mean What You Think](https://mlsanity.com/paper/2503.03150/) (2025)
- [Cultural Evolution of Cooperation among LLM Agents](https://mlsanity.com/paper/2412.10270/) (2024)
- [Nepotistically Trained Generative-AI Models Collapse](https://mlsanity.com/paper/2311.12202/) (2023)
- [Social Network Structure Shapes Innovation: Experience-sharing in RL with SAPIENS](https://mlsanity.com/paper/2206.05060/) (2022)
- [Artificial Generational Intelligence: Cultural Accumulation in Reinforcement Learning](https://mlsanity.com/paper/2406.00392/) (2024)

## Built on by (most cited first, 2 of 2)

- [Beneath the Diff: Diagnosing and Mitigating Algorithmic Mode Collapse in Code-Level Autonomous Research Loops](https://mlsanity.com/paper/2609.00077/) (2026)
- [The Anti-Ouroboros Effect: Emergent Resilience in Large Language Models from Recursive Selective Feedback](https://mlsanity.com/paper/2509.10509/) (2025)

## Abstract (arXiv)

Large language models (LLMs) are increasingly used in the creation of online content, creating feedback loops as subsequent generations of models will be trained on this synthetic data. Such loops were shown to lead to distribution shifts - models misrepresenting the true underlying distributions of human data (also called model collapse). However, how human data properties affect such shifts remains poorly understood. In this paper, we provide the first empirical examination of the effect of such properties on the outcome of recursive training. We first confirm that using different human datasets leads to distribution shifts of different magnitudes. Through exhaustive manipulation of dataset properties combined with regression analyses, we then identify a set of properties predicting distribution shift magnitudes. Lexical diversity is found to amplify these shifts, while semantic diversity and data quality mitigate them. Furthermore, we find that these influences are highly modular: data scrapped from a given internet domain has little influence on the content generated for another domain. Finally, experiments on political bias reveal that human data properties affect whether the initial bias will be amplified or reduced. Overall, our results portray a novel view, where different parts of internet may undergo different types of distribution shift.