Unverified50% confidenceOpinionExact time
Traditional unlearning methods often use gradient ascent or preference optimization to directly suppress the probability of target knowledge; the problem is that they only tear down without building up—the model learns not to say it, but does not learn what it should say.
1
Sources
50%
Confidence
Long-term
Relevance
11/17/2026
First Seen
Sources
Related Entities
Related Claims
VerifiedThe core challenge of internalizing knowledge into model parameters is the Catastrophic Forgetting problem, which is a difficulty not yet fully resolved in the field of continual learning.80% similarUnverifiedWhen the contribution of fine-tuning is artificially reduced, the advantage brought by cross-prompt inference also shrinks substantially, supporting at the causal level the judgment that rare-class knowledge is still retained in the pretrained part.77% similarUnverifiedA model's ability to comply with negative instructions—such as ignoring certain information—may be far weaker than assumed.77% similarUnverifiedAdversarial examples are inputs that have been deliberately fine-tuned to be almost imperceptible to humans yet can induce machine learning models to make incorrect judgments77% similarUnverified承认不知道并展示如何推理,远胜于强行编造标准答案,面试官更看重结构化解决问题能力而非记忆力76% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1179615API
curl https://kongchang.com/api/v1/knowledge/claims/1179615MCP
get_claim(id=1179615)