# Understanding Catastrophic Forgetting in Language Models via Implicit Inference

arXiv: https://arxiv.org/abs/2309.10105 · HTML: https://mlsanity.com/paper/2309.10105/

## Key facts

- Cited by 144 papers in the map (data through 14 Oct 2026).
- Placed in Reward hacking and misspecification, one of 10 directions in LLM alignment / RLHF / preference optimisation.
- Builds on 48 of its 74 references that are in the map.
- First posted on arXiv 18 Sep 2023.

## Placed in

- LLM alignment / RLHF / preference optimisation / [Reward hacking and misspecification](https://mlsanity.com/alignment/reward-hacking-and-misspecification/)

## Builds on (most cited first, 30 of 48)

- [Attention Is All You Need](https://mlsanity.com/paper/1706.03762/) (2017)
- [Training language models to follow instructions with human feedback](https://mlsanity.com/paper/2203.02155/) (2022)
- [Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer](https://mlsanity.com/paper/1910.10683/) (2019)
- [LLaMA: Open and Efficient Foundation Language Models](https://mlsanity.com/paper/2302.13971/) (2023)
- [Llama 2: Open Foundation and Fine-Tuned Chat Models](https://mlsanity.com/paper/2307.09288/) (2023)
- [GLUE: A Multi-Task Benchmark and Analysis Platform for Natural Language Understanding](https://mlsanity.com/paper/1804.07461/) (2018)
- [Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networks](https://mlsanity.com/paper/1703.03400/) (2017)
- [Overcoming catastrophic forgetting in neural networks](https://mlsanity.com/paper/1612.00796/) (2016)
- [Deep reinforcement learning from human preferences](https://mlsanity.com/paper/1706.03741/) (2017)
- [Finetuned Language Models Are Zero-Shot Learners](https://mlsanity.com/paper/2109.01652/) (2021)
- [OPT: Open Pre-trained Transformer Language Models](https://mlsanity.com/paper/2205.01068/) (2022)
- [Universal and Transferable Adversarial Attacks on Aligned Language Models](https://mlsanity.com/paper/2307.15043/) (2023)
- [A Broad-Coverage Challenge Corpus for Sentence Understanding through Inference](https://mlsanity.com/paper/1704.05426/) (2017)
- [Learning to summarize from human feedback](https://mlsanity.com/paper/2009.01325/) (2020)
- [The Pile: An 800GB Dataset of Diverse Text for Language Modeling](https://mlsanity.com/paper/2101.00027/) (2020)
- [Fine-Tuning Language Models from Human Preferences](https://mlsanity.com/paper/1909.08593/) (2019)
- [Jailbroken: How Does LLM Safety Training Fail?](https://mlsanity.com/paper/2307.02483/) (2023)
- [Google's Neural Machine Translation System: Bridging the Gap between Human and Machine Translation](https://mlsanity.com/paper/1609.08144/) (2016)
- [Multitask Prompted Training Enables Zero-Shot Task Generalization](https://mlsanity.com/paper/2110.08207/) (2021)
- [Rethinking the Role of Demonstrations: What Makes In-Context Learning Work?](https://mlsanity.com/paper/2202.12837/) (2022)
- [Continual Lifelong Learning with Neural Networks: A Review](https://mlsanity.com/paper/1802.07569/) (2018)
- [Making Pre-trained Language Models Better Few-shot Learners](https://mlsanity.com/paper/2012.15723/) (2020)
- [An Explanation of In-context Learning as Implicit Bayesian Inference](https://mlsanity.com/paper/2111.02080/) (2021)
- [XNLI: Evaluating Cross-lingual Sentence Representations](https://mlsanity.com/paper/1809.05053/) (2018)
- [Transformers learn in-context by gradient descent](https://mlsanity.com/paper/2212.07677/) (2022)
- [An Empirical Investigation of Catastrophic Forgetting in Gradient-Based Neural Networks](https://mlsanity.com/paper/1312.6211/) (2013)
- [What Can Transformers Learn In-Context? A Case Study of Simple Function Classes](https://mlsanity.com/paper/2208.01066/) (2022)
- [Cross-Task Generalization via Natural Language Crowdsourcing Instructions](https://mlsanity.com/paper/2104.08773/) (2021)
- [What learning algorithm is in-context learning? Investigations with linear models](https://mlsanity.com/paper/2211.15661/) (2022)
- [An Empirical Study of Catastrophic Forgetting in Large Language Models During Continual Fine-tuning](https://mlsanity.com/paper/2308.08747/) (2023)

## Built on by (most cited first, 30 of 144)

- [A Survey on Knowledge Distillation of Large Language Models](https://mlsanity.com/paper/2402.13116/) (2024)
- [Survey on Factuality in Large Language Models: Knowledge, Retrieval and Domain-Specificity](https://mlsanity.com/paper/2310.07521/) (2023)
- [Latent Adversarial Training Improves Robustness to Persistent Harmful Behaviors in LLMs](https://mlsanity.com/paper/2407.15549/) (2024)
- [Eight Methods to Evaluate Robust Unlearning in LLMs](https://mlsanity.com/paper/2402.16835/) (2024)
- [Continual Learning for Large Language Models: A Survey](https://mlsanity.com/paper/2402.01364/) (2024)
- [Empowering Biomedical Discovery with AI Agents](https://mlsanity.com/paper/2404.02831/) (2024)
- [MLLM-Protector: Ensuring MLLM's Safety without Hurting Performance](https://mlsanity.com/paper/2401.02906/) (2024)
- [Mechanistically analyzing the effects of fine-tuning on procedurally defined tasks](https://mlsanity.com/paper/2311.12786/) (2023)
- [Guardrail Baselines for Unlearning in LLMs](https://mlsanity.com/paper/2403.03329/) (2024)
- [Retaining by Doing: The Role of On-Policy Data in Mitigating Forgetting](https://mlsanity.com/paper/2510.18874/) (2025)
- [Defending Against Unforeseen Failure Modes with Latent Adversarial Training](https://mlsanity.com/paper/2403.05030/) (2024)
- [Controlling Language and Diffusion Models by Transporting Activations](https://mlsanity.com/paper/2410.23054/) (2024)
- [Refusal-Trained LLMs Are Easily Jailbroken As Browser Agents](https://mlsanity.com/paper/2410.13886/) (2024)
- [PolyGuard: A Multilingual Safety Moderation Tool for 17 Languages](https://mlsanity.com/paper/2504.04377/) (2025)
- [What Makes and Breaks Safety Fine-tuning? A Mechanistic Study](https://mlsanity.com/paper/2407.10264/) (2024)
- [The Best Instruction-Tuning Data are Those That Fit](https://mlsanity.com/paper/2502.04194/) (2025)
- [A Safe Harbor for AI Evaluation and Red Teaming](https://mlsanity.com/paper/2403.04893/) (2024)
- [Multi-Attribute Steering of Language Models via Targeted Intervention](https://mlsanity.com/paper/2502.12446/) (2025)
- [Crosslingual Reasoning through Test-Time Scaling](https://mlsanity.com/paper/2505.05408/) (2025)
- [Personalized Large Language Models](https://mlsanity.com/paper/2402.09269/) (2024)
- [A Comprehensive Study of Multimodal Large Language Models for Image Quality Assessment](https://mlsanity.com/paper/2403.10854/) (2024)
- [HtmlRAG: HTML is Better Than Plain Text for Modeling Retrieved Knowledge in RAG Systems](https://mlsanity.com/paper/2411.02959/) (2024)
- [Reinforcement Learning Meets Large Language Models: A Survey of Advancements and Applications Across the LLM Lifecycle](https://mlsanity.com/paper/2509.16679/) (2025)
- [Small Models, Big Insights: Leveraging Slim Proxy Models To Decide When and What to Retrieve for LLMs](https://mlsanity.com/paper/2402.12052/) (2024)
- [AI Safety in Generative AI Large Language Models: A Survey](https://mlsanity.com/paper/2407.18369/) (2024)
- [Reflect, Retry, Reward: Self-Improving LLMs via Reinforcement Learning](https://mlsanity.com/paper/2505.24726/) (2025)
- [HSplitLoRA: A Heterogeneous Split Parameter-Efficient Fine-Tuning Framework for Large Language Models](https://mlsanity.com/paper/2505.02795/) (2025)
- [RL Fine-Tuning Heals OOD Forgetting in SFT](https://mlsanity.com/paper/2509.12235/) (2025)
- [Self-MoE: Towards Compositional Large Language Models with Self-Specialized Experts](https://mlsanity.com/paper/2406.12034/) (2024)
- [PolygloToxicityPrompts: Multilingual Evaluation of Neural Toxic Degeneration in Large Language Models](https://mlsanity.com/paper/2405.09373/) (2024)

## Abstract (arXiv)

We lack a systematic understanding of the effects of fine-tuning (via methods such as instruction-tuning or reinforcement learning from human feedback), particularly on tasks outside the narrow fine-tuning distribution. In a simplified scenario, we demonstrate that improving performance on tasks within the fine-tuning data distribution comes at the expense of capabilities on other tasks. We hypothesize that language models implicitly infer the task of the prompt and that fine-tuning skews this inference towards tasks in the fine-tuning distribution. To test this, we propose Conjugate Prompting, which artificially makes the task look farther from the fine-tuning distribution while requiring the same capability, and we find that this recovers some of the pretraining capabilities in our synthetic setup. Since real-world fine-tuning distributions are predominantly English, we apply conjugate prompting to recover pretrained capabilities in LLMs by simply translating the prompts to different languages. This allows us to recover in-context learning abilities lost via instruction tuning, natural reasoning capability lost during code fine-tuning, and, more concerningly, harmful content generation suppressed by safety fine-tuning in chatbots like ChatGPT.