Unverified50% confidenceFactExact time
When the execution agent has access to the Wiki layer, it directly retrieves answers, bypassing procedural skills, which leads to shortcut learning and contaminates the execution trajectories.
1
Sources
50%
Confidence
Long-term
Relevance
8/31/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedAblation experiments showed that allowing the reasoning agent to read the Wiki layer during training leads to a drop in win rate78% similarUnverifiedThe architectural principle given by WikiSkill is that the Wiki layer should be visible only to the skill-proposing agent and not to the execution agent.76% similarUnverified通过系统提示词约束智能体只回答知识库中有的问题，可有效防止智能体回答无关问题并引导转人工66% similarUnverifiedUnlike supervised learning, which relies on labeled data, reinforcement learning does not require humans to provide correct answers; instead, it allows agents to autonomously discover optimal behavioral paths through trial and error.65% similarUnverifiedAI models capable of vulnerability discovery may allow people without deep security knowledge to find and exploit vulnerabilities, lowering the barrier to attacks.65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/830037API
curl https://kongchang.com/api/v1/knowledge/claims/830037MCP
get_claim(id=830037)