Unverified50% confidenceSolutionExact time
The developers corrected the reset distribution via forward kinematics, rejected invalid configurations, and separated recovery episodes from standing initialization to keep the reward signal from being polluted by "cheating" behaviors.
1
Sources
50%
Confidence
Long-term
Relevance
10/17/2026
First Seen
Sources
Training Whole-Body Recovery Policies for Humanoid Robots: From Reward Design to DC Motor Modeling
redditr/reinforcementlearning10/17/2026
Related Entities
Related Claims
UnverifiedThe Abliteration technique identifies and suppresses the direction vector responsible for refusal behavior in the model's residual stream, using orthogonal projection to remove that direction from the weight matrices, thereby eliminating refusal behavior without retraining.67% similarUnverifiedAfter a gate is triggered, developers have three options: Resume, Promote (to 100%), and Roll back.65% similarUnverifiedThe triggers for silent drift often come from external sources, such as version upgrades of the underlying large model, fine-tuning of API behavior, or changes to tool interfaces—modifications that are not in the codebase yet directly alter Agent behavior.62% similarUnverifiedIncluding tasks where reframing is harmful simultaneously allows genuine reframing ability to be decoupled from systematic reframing bias at the score level.62% similarUnverified重启应用与系统可清除临时的渲染异常状态，是排查布局问题的基础方法62% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1052192API
curl https://kongchang.com/api/v1/knowledge/claims/1052192MCP
get_claim(id=1052192)