Unverified50% confidenceOpinionExact time
Research shows that multimodal large models surpass their ability to detect fake news in their ability to generate fake news, meaning attacker capabilities are growing faster than defender capabilities.
1
Sources
50%
Confidence
Long-term
Relevance
9/30/2026
First Seen
Sources
Related Entities
Related Claims
Unverified自动化对抗样本生成面临覆盖率幻觉挑战，自动生成的攻击样本可能数量庞大但高度同质化，遗漏新颖攻击路径64% similarUnverified模型压缩技术（量化、剪枝、知识蒸馏）会不成比例地损害模型的事实准确性，量化后模型在事实性问答任务上的幻觉率可能显著上升63% similarUnverifiedCovert, persistent intrusions are harder for security teams to detect than one-time data breaches, and the scope of victims continues to expand over time.63% similarUnverifiedDiCE生成反事实解释计算开销较大，对全量样本生成反事实在大规模生产数据上可能成为性能瓶颈63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/961215API
curl https://kongchang.com/api/v1/knowledge/claims/961215MCP
get_claim(id=961215)