[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"$fA9j__zZ6Cm9qV51TrC62zHCqJkRy-F16wPvpifqBin4":3},{"slug":4,"term":5,"shortDefinition":6,"seoTitle":7,"seoDescription":8,"explanation":9,"relatedTerms":10,"faq":20,"category":27},"catastrophic-forgetting","Catastrophic Forgetting","A phenomenon where fine-tuning a model on new data causes it to lose previously learned knowledge and capabilities.","Catastrophic Forgetting in llm - InsertChat","Learn what catastrophic forgetting is, why it happens during fine-tuning, and how to prevent knowledge loss in LLMs.","Catastrophic Forgetting matters in llm work because it changes how teams evaluate quality, risk, and operating discipline once an AI system leaves the whiteboard and starts handling real traffic. A strong page should therefore explain not only the definition, but also the workflow trade-offs, implementation choices, and practical signals that show whether Catastrophic Forgetting is helping or creating new failure modes. Catastrophic forgetting occurs when fine-tuning a pre-trained model on new data causes it to lose knowledge and capabilities acquired during pre-training. The model overwrites previously learned representations with new ones, degrading performance on tasks outside the fine-tuning domain.\n\nThis happens because gradient descent updates to the model weights are not constrained to preserve old knowledge while learning new tasks. Without precautions, the model fully adapts to the new data distribution, losing its general capabilities. A model fine-tuned aggressively on medical texts might forget how to write code or engage in casual conversation.\n\nMitigation strategies include using low learning rates (preserving most pre-trained weights), parameter-efficient fine-tuning (modifying only a small subset of parameters via LoRA), mixing in general-purpose data during fine-tuning, early stopping (training for fewer steps), and regularization techniques that penalize large deviations from pre-trained weights.\n\nCatastrophic Forgetting is often easier to understand when you stop treating it as a dictionary entry and start looking at the operational question it answers. Teams normally encounter the term when they are deciding how to improve quality, lower risk, or make an AI workflow easier to manage after launch.\n\nThat is also why Catastrophic Forgetting gets compared with Fine-Tuning, LoRA, and Continued Pre-training. The overlap can be real, but the practical difference usually sits in which part of the system changes once the concept is applied and which trade-off the team is willing to make.\n\nA useful explanation therefore needs to connect Catastrophic Forgetting back to deployment choices. When the concept is framed in workflow terms, people can decide whether it belongs in their current system, whether it solves the right problem, and what it would change if they implemented it seriously.\n\nCatastrophic Forgetting also tends to show up when teams are debugging disappointing outcomes in production. The concept gives them a way to explain why a system behaves the way it does, which options are still open, and where a smarter intervention would actually move the quality needle instead of creating more complexity.",[11,14,17],{"slug":12,"name":13},"continual-learning","Continual Learning",{"slug":15,"name":16},"fine-tuning","Fine-Tuning",{"slug":18,"name":19},"lora","LoRA",[21,24],{"question":22,"answer":23},"How do I know if my model is experiencing catastrophic forgetting?","Evaluate the model on general benchmarks (MMLU, HellaSwag, etc.) before and after fine-tuning. Significant degradation indicates forgetting. Also test on tasks the model could handle before fine-tuning to check for capability loss. Catastrophic Forgetting becomes easier to evaluate when you look at the workflow around it rather than the label alone. In most teams, the concept matters because it changes answer quality, operator confidence, or the amount of cleanup that still lands on a human after the first automated response.",{"question":25,"answer":26},"Does LoRA prevent catastrophic forgetting?","LoRA significantly reduces catastrophic forgetting because it only modifies a small number of additional parameters while keeping the base model frozen. The original knowledge is preserved in the frozen weights. This is one of the main advantages of parameter-efficient methods. That practical framing is why teams compare Catastrophic Forgetting with Fine-Tuning, LoRA, and Continued Pre-training instead of memorizing definitions in isolation. The useful question is which trade-off the concept changes in production and how that trade-off shows up once the system is live.","llm"]