# Talking to Yourself: Defying Forgetting in Large Language Models

arXiv: https://arxiv.org/abs/2602.20162 · HTML: https://mlsanity.com/paper/2602.20162/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Placed in Test-time compute and efficient thinking, one of 17 directions in LLM reasoning.
- Builds on 22 of its 36 references that are in the map.
- First posted on arXiv 23 Jan 2026.

## Placed in

- LLM reasoning / [Test-time compute and efficient thinking](https://mlsanity.com/lr/test-time-compute-and-efficient-thinking/)

## Builds on (most cited first, 22 of 22)

- [The Llama 3 Herd of Models](https://mlsanity.com/paper/2407.21783/) (2024)
- [Training Verifiers to Solve Math Word Problems](https://mlsanity.com/paper/2110.14168/) (2021)
- [Measuring Massive Multitask Language Understanding](https://mlsanity.com/paper/2009.03300/) (2020)
- [LlamaFactory: Unified Efficient Fine-Tuning of 100+ Language Models](https://mlsanity.com/paper/2403.13372/) (2024)
- [Super-NaturalInstructions: Generalization via Declarative Instructions on 1600+ NLP Tasks](https://mlsanity.com/paper/2204.07705/) (2022)
- [AGIEval: A Human-Centric Benchmark for Evaluating Foundation Models](https://mlsanity.com/paper/2304.06364/) (2023)
- [An Empirical Study of Catastrophic Forgetting in Large Language Models During Continual Fine-tuning](https://mlsanity.com/paper/2308.08747/) (2023)
- [MAmmoTH: Building Math Generalist Models through Hybrid Instruction Tuning](https://mlsanity.com/paper/2309.05653/) (2023)
- [Measuring Catastrophic Forgetting in Neural Networks](https://mlsanity.com/paper/1708.02072/) (2017)
- [Magpie: Alignment Data Synthesis from Scratch by Prompting Aligned LLMs with Nothing](https://mlsanity.com/paper/2406.08464/) (2024)
- [Using an LLM to Help With Code Understanding](https://mlsanity.com/paper/2307.08177/) (2023)
- [Episodic Memory in Lifelong Language Learning](https://mlsanity.com/paper/1906.01076/) (2019)
- [Language Models Meet World Models: Embodied Experiences Enhance Language Models](https://mlsanity.com/paper/2305.10626/) (2023)
- [Mitigating Catastrophic Forgetting in Large Language Models with Self-Synthesized Rehearsal](https://mlsanity.com/paper/2403.01244/) (2024)
- [Spurious Forgetting in Continual Learning of Language Models](https://mlsanity.com/paper/2501.13453/) (2025)
- [A Survey of Mix-based Data Augmentation: Taxonomy, Methods, Applications, and Explainability](https://mlsanity.com/paper/2212.10888/) (2022)
- [Towards Automatic Generation of Shareable Synthetic Clinical Notes Using Neural Language Models](https://mlsanity.com/paper/1905.07002/) (2019)
- [How to Alleviate Catastrophic Forgetting in LLMs Finetuning? Hierarchical Layer-Wise and Element-Wise Regularization](https://mlsanity.com/paper/2501.13669/) (2025)
- [Improved Supervised Fine-Tuning for Large Language Models to Mitigate Catastrophic Forgetting](https://mlsanity.com/paper/2506.09428/) (2025)
- [The Self-Improvement Paradox: Can Language Models Bootstrap Reasoning Capabilities without External Scaffolding?](https://mlsanity.com/paper/2502.13441/) (2025)
- [Multilingual Mathematical Reasoning: Advancing Open-Source LLMs in Hindi and English](https://mlsanity.com/paper/2412.18415/) (2024)
- [Preserving Knowledge in Large Language Model with Model-Agnostic Self-Decompression](https://mlsanity.com/paper/2406.11354/) (2024)

## Built on by (most cited first, 1 of 1)

- [Crafting Reversible SFT Behaviors in Large Language Models](https://mlsanity.com/paper/2605.06632/) (2026)

## Abstract (arXiv)

Catastrophic forgetting remains a major challenge when fine-tuning large language models (LLMs) on narrow, task-specific data, often degrading their general knowledge and reasoning abilities. We propose SA-SFT, a lightweight self-augmentation routine in which an LLM generates self-dialogues prior to fine-tuning, and the resulting self-authored data are mixed with task data without modifying optimization or training schedules. Despite requiring no external data or additional tuning, SA-SFT consistently mitigates catastrophic forgetting while improving in-domain performance. Across 50 evaluation scenarios, it maintains performance comparable to the original model and achieves the best results in 40 cases, outperforming common baselines such as layer freezing and external data mixing. Guided by these empirical findings, we further present a theoretical analysis suggesting that forgetting can partly stem from style-induced parameter drift, and that self-alignment through self-generated data provides an effective means to counteract this effect. Overall, our results indicate that self-augmentation offers a simple and effective mechanism for robust LLM adaptation without incurring catastrophic forgetting.