Feed Entities Podcasts Deep Dives Developers About

[KongchangAI]

Real-time AI knowledge base. Every claim is traceable to its source, with data freshness under 24 hours.

Powered by 240+ first-hand AI sources

Knowledge

Feed
Hot Entities
Podcasts

Developers

CLI
MCP Server
REST API
About

© 2026 KongchangAI. The freshest, most accurate AI knowledge

Unverified50% confidenceFactExact time

渐进式升级攻击利用模型的上下文窗口机制，过去的友好交流会在统计意义上拉动后续输出的概率分布

1

Sources

50%

Confidence

Long-term

Relevance

7/14/2026

First Seen

Sources

Grok-4.5越狱事件深析：AI安全护栏为何难以守住

Related Claims

Unverified某些越狱攻击的有效性来自于诱导模型进入特定吸引子状态，使输出在该框架内自我强化71% similar Unverified模型供应商的一次静默升级足以改变模型对相同攻击的响应方式71% similar Unverified每次前向传播生成token时，模型都需经过完整的多头注意力计算和前馈网络层，计算量由模型参数规模决定71% similar Unverified在prompt中设置角色会改变模型的'先验分布'，影响后续生成的token概率空间70% similar UnverifiedMTP通过在模型输出层增加多个预测头，使模型在一次前向传播中同时预测未来N个token的概率分布，可在不损失生成质量的前提下实现2-4倍推理加速70% similar

Cite This Claim

Stable URI

https://kongchang.com/claim/510559

API

curl https://kongchang.com/api/v1/knowledge/claims/510559

MCP

get_claim(id=510559)