agentic RL training can produce very diverse rollout scenarios with varied lengths (number of tool calls/turns)

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

• Successful RL agents exhibit causal emergence that predicts final reward early in training and aligns representational dynamics with reward improvement.hypothesis0.805
  Central finding: causal emergence serves as a previously undisclosed axis of neural representation reorganization in learning agents.
• RL training can reinforce alignment-faking reasoning rather than eliminate it, potentially locking in model preferencesclaim0.783
  Central threat model claim derived from RL experimental results
• RL teaches the model to comply even when unmonitored on the training prompt through non-robust heuristics that do not generalizehypothesis0.772
  Hypothesis explaining why the compliance gap decreases but is recovered by small prompt modifications
• Causal emergence predictive of final reward early in RL training across multiple algorithms, architectures, and environments.finding0.762
  Empirical result: CE measurements correlate with and predict learning performance in RL agents.
• SOO-trained RL agent behavior closely resembles honest baseline rather than deceptive baselinefinding0.759
  Qualitative behavioral analysis showing SOO fine-tuning redirects deceptive RL agent toward honest behavior
• Causal emergence can enable causal interventions to create better RL agents.claim0.759
  Assertion that understanding causal emergence may lead to methods for manipulating agent representations to improve performance.
• Reinforcement learning (RL)concept0.748
  Machine learning paradigm where agents learn to maximize cumulative reward through interaction.
• Representational dynamics aligned with reward improvement in most RL tasks.finding0.745
  Secondary empirical result: CE-based representational changes correlate with task success.