Unverified50% confidenceSolutionExact time
During the RL phase, Iris deploys the reward judge and observation summarizer directly inside the training cluster to improve training efficiency.
1
Sources
50%
Confidence
Long-term
Relevance
9/12/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedIris adopts an SFT-RL climbing strategy, alternating between supervised fine-tuning and reinforcement learning, where high-quality behavioral trajectories discovered during the RL phase are fed back to the SFT phase as new supervision signals.74% similarUnverifiedA high-quality RL training environment must satisfy three conditions: providing a task distribution close to real-world scenarios, delivering accurate and discriminative reward signals, and supporting large-scale parallel sampling to accelerate training.71% similarUnverifiedThe higher priority of system prompts over user instructions is deliberately reinforced through dedicated alignment training data during the RLHF stage.67% similarUnverifiedDuring RL training, more refined reward mechanisms or adversarial supervision can be introduced to detect and penalize suspicious encoding behavior.67% similarUnverifiedRL训练涉及环境交互（CPU密集）和策略优化（GPU密集）两个阶段的交替66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/915228API
curl https://kongchang.com/api/v1/knowledge/claims/915228MCP
get_claim(id=915228)