Unverified50% confidenceFactExact time
SFT has a structural contradiction of catastrophic forgetting: the more deeply new data is learned, the more original capabilities are forgotten.
1
Sources
50%
Confidence
Long-term
Relevance
10/16/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedGradient-based methods often overwrite old knowledge when learning new tasks, causing catastrophic forgetting, whereas population-based evolutionary systems naturally maintain an archive of solution diversity.70% similarUnverifiedCost reduction often comes with quality trade-offs: shorter context may cause the model to miss critical information, and more aggressive caching may introduce stale data.69% similarUnverified用新数据更新模型时可能破坏原有能力，即灾难性遗忘，工业界通过失败案例库、数据增强与混合回放技术来整合新数据67% similarVerifiedIf only the data of newly added classes is used to fine-tune an expanded classification head, the model is highly prone to catastrophic forgetting, rapidly losing its ability to recognize the original classes.67% similarUnverifiedRNNs face the vanishing gradient problem, whose essence is the cascading multiplication effect under the chain rule, making it difficult to learn long-range historical information.66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1041747API
curl https://kongchang.com/api/v1/knowledge/claims/1041747MCP
get_claim(id=1041747)