Unverified50% confidenceFactExact time
The algorithm uses a PPO plus LSTM policy with about 1.5 million to 1.65 million parameters; the action space includes move/stop, basic attack, QWER, Flash and Barrier, with a pointer head for target selection, and illegal actions are masked.
1
Sources
50%
Confidence
Long-term
Relevance
11/2/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedAction masking sets the probabilities of illegal actions to zero and renormalizes when the policy network outputs action probabilities, ensuring that the agent samples only from legal actions; it can significantly accelerate training but may introduce implicit preferences leading to path dependence.68% similarUnverified基于策略梯度的算法如PPO和GRPO能够处理离散与连续混合的动作空间67% similarUnverifiedThe runtime must set hard interception at points such as tool calls, path access and action triggering in order to limit the impact of prompt injection to a controllable range.65% similarUnverified基于GCG（贪婪坐标梯度）算法的自动化越狱可在单张消费级GPU上、数小时内生成数千条有效攻击提示65% similarVerifiedPPO (Proximal Policy Optimization) limits the step size of each policy update through a clipped objective function, and has become the de facto standard for continuous control tasks.63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1125550API
curl https://kongchang.com/api/v1/knowledge/claims/1125550MCP
get_claim(id=1125550)