Reinforcement-Learning

On the Optimal Teacher for On-Policy (Self-)Distillation
A short derivation of the teacher objective induced by asking OPD to match a plain RL policy-gradient update.
Randy Ardywibowo
Jul 9, 2026 5 min read
Little's Law and Resource Allocations for Reinforcement Learning
Reinforcement Learning (RL) consists of generating rollouts and training on them. These two systems are separate and can be largely disaggregated to improve throughput. As such, we can scale each component separately.
Randy Ardywibowo
May 28, 2026 9 min read
End-to-End Multi-Agent Reinforcement Learning
A trace-level view of reinforcement learning for multi-agent, tool-using, and compacted language-model systems.
Randy Ardywibowo
Apr 27, 2026 5 min read