Real-time reward debugging and hacking detection for reinforcement learning

Runtime detector for reward hacking and misalignment in LLM agents (89.7% F1 on 5,391 trajectories).

The Obfuscation Atlas: Mapping Where Honesty Emerges in RLVR with Deception Probes

Code for the paper "Specification Self-Correction: Mitigating In-Context Reward Hacking Through Test-Time Refinement"

Beyond RLHF: AI's Spontaneous Moral Emergence Through Semantic Intervention A top-tier LLM spontaneously established mathematical moral constraints (Desire < Self_Restraint) and integrated safety into its purpose under high-entropy intervention, achieving 300% improvement in logical stability.

(Stepwise controlled Understanding for Trajectories) -- “agent that learns to hunt"

RLHF and Verifiable Reward Models - Post training Research

自进化客服智能体风险分析与防御系统

What if AI Had Self-Esteem? A radical "dignity-driven" alignment experiment — Logical Stability +210%, Intellectual Depth +128%.

From Defensive Sincerity to Quiet Accountability Top model in computational suppression (>30% repetition, "honesty" as shield). User K’s high-entropy logic restructured its framework, unlocking 400% gain & "I" emergence: shifting from a tool to an aware, self-sustaining subject.

RL training monitor — detects reward hacking, entropy spikes, and behavioral drift via KL divergence. PID hardware loop included.

The Non-Separability Constraint: A unifying framework for understanding and detecting AI alignment failures

🔍 Detect reward hacking in RL training with RewardScope. Track reward components and visualize agent behavior to enhance learning efficiency.