{"id":"W6929501122","doi":"10.48550/arxiv.1711.03087","title":"Exploration in NetHack With Secret Discovery","year":2017,"lang":"en","type":"preprint","venue":"arXiv (Cornell University)","topic":"Artificial Intelligence in Games","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"McGill University","funders":"","keywords":"Heuristics; Automation; Point (geometry); Greedy algorithm; Feature (linguistics); Face (sociological concept); Key (lock)","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.0002334293,0.0002640075,0.0002763567,0.0002381058,0.0001411311,0.0005005529,0.002386559,0.0002202309,0.000009087164],"category_scores_gemma":[0.0000470945,0.000272825,0.0000863036,0.0002679749,0.0001912207,0.002653965,0.00160032,0.0005401627,0.00009593459],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001892201,"about_ca_system_score_gemma":0.0002670228,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0007415262,"about_ca_topic_score_gemma":0.001602117,"domain_scores_codex":[0.998274,0.0001109276,0.0001809624,0.001013752,0.0001094566,0.0003108965],"domain_scores_gemma":[0.9976418,0.00007845275,0.0003131664,0.00177288,0.0001062922,0.0000874114],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00006905921,0.00009886982,0.01665407,0.00004797738,0.00004108381,0.0008908386,0.001603686,0.7254314,0.00002709117,0.2525183,0.0001764174,0.002441269],"study_design_scores_gemma":[0.0001982985,0.0001027432,0.002619907,0.0003241306,0.00002521168,0.00000398372,0.0003606859,0.7258418,0.001129856,0.2682243,0.0004564075,0.0007126738],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.2858442,0.00002233192,0.7101841,0.0002006086,0.0003819497,0.0002276019,0.00000408406,0.000104279,0.003030895],"genre_scores_gemma":[0.9961256,0.0001345171,0.001481752,0.00004574022,0.00006928622,0.000001938704,0.00001017097,0.00001463144,0.00211634],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.7102814,"threshold_uncertainty_score":0.9999724,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1523072562991754,"score_gpt":0.2247996867509447,"score_spread":0.07249243045176929,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}