Imitation Dominates Reinforcement: Direct In-Context RL Is Closer to ICL Than RL

Connections are not prepared for this record yet, or its source metadata has changed. The original record remains available while background snapshots are rebuilt.