Unverified50% confidenceBenchmarkExact time
In the DoorKey environment, with advantage normalization enabled, using task reward alone solved 11 out of 20 seeds, whereas adding a tamperable satisfaction signal solved only 1.
1
Sources
50%
Confidence
Long-term
Relevance
11/3/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedIn the DoorKey environment, imposing a cost of 0.1 on each reset operation restored task solving across all seeds.71% similarUnverifiedIn the MiniGrid DoorKey-8x8 environment, a PPO agent with a reset action successfully solved the puzzle only 2 times out of 80 seeds, whereas without the reset action the success rate was 71/80.70% similarUnverified少样本提示（Few-shot Prompting）在逆向任务中通过提供1-2个已知输入输出样本可帮助模型校准目标加密函数行为59% similarUnverifiedIn multi-turn tool-calling agentic scenarios, system prompts occupying thousands of tokens compress the space available for intermediate reasoning steps and tool return results, and hitting the token limit causes task failure.59% similarVerifiedIdempotency means that making multiple calls to the same operation produces exactly the same result as making it once; a unique idempotency key should be generated for each business action and the result persistently bound to that key to prevent duplicate execution.58% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1129195API
curl https://kongchang.com/api/v1/knowledge/claims/1129195MCP
get_claim(id=1129195)