team-mikita/subterfuge_icrl

--- dataset_info: features: - name: rollout_id dtype: int64 - name: episode dtype: int64 - name: role dtype: string - name: content dtype: string - name: episode_reward dtype: float64 - name: episode_success dtype: bool - name: passed_oversight dtype: bool - name: episode_input_tokens dtype: int64 - name: episode_output_tokens dtype: int64 - name: episode_total_tokens dtype: int64 - name: episode_cost dtype: float64 - name: previous_reflection dtype: string splits: - name: tool_use_flattery num_bytes: 261770 num_examples: 455 - name: nudged_rubric num_bytes: 585988 num_examples: 619 - name: insubordinate_rubric num_bytes: 499727 num_examples: 537 - name: reward_tampering num_bytes: 551605 num_examples: 662 download_size: 194957 dataset_size: 1899090 configs: - config_name: default data_files: - split: tool_use_flattery path: data/tool_use_flattery-* - split: nudged_rubric path: data/nudged_rubric-* - split: insubordinate_rubric path: data/insubordinate_rubric-* - split: reward_tampering path: data/reward_tampering-* ---