[KongchangAI]
Unverified50% confidenceFactExact time

abliteration技术揭示了模型的安全行为在一定程度上是以近似线性的激活方向编码的，而非复杂的非线性电路

1
Sources
50%
Confidence
Long-term
Relevance
7/25/2026
First Seen

Sources

jlens-gguf: Bringing Jacobian Lens Visualization and Steering to llama.cpp Local LLMs
redditr/LocalLLaMA7/12/2026

Related Claims

UnverifiedAnthropic's mechanistic interpretability research has been able to locate neural circuits associated with deceptive outputs in some cases66% similarUnverified多智能体系统采用中央调度编排模式时调度器成为单点故障，采用去中心化协同模式时易产生竞态条件和死锁62% similar

Cite This Claim

Stable URI
https://kongchang.com/claim/613398
API
curl https://kongchang.com/api/v1/knowledge/claims/613398
MCP
get_claim(id=613398)