{"id":"W2756674674","doi":"10.1609/aiide.v13i1.12923","title":"Learning Combat in NetHack","year":2021,"lang":"en","type":"article","venue":"Proceedings of the AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment","topic":"Artificial Intelligence in Games","field":"Computer Science","cited_by":3,"is_retracted":false,"has_abstract":true,"ca_institutions":"McGill University","funders":"","keywords":"Scripting language; Computer science; Automation; Adversary; Variety (cybernetics); Point (geometry); Action (physics); Artificial intelligence; Human–computer interaction; Space (punctuation); Software engineering; Machine learning; Programming language; Computer security; Engineering","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":[],"consensus_categories":[],"category_scores_codex":[0.0002425459,0.0002687004,0.0003199316,0.0001482524,0.0001217331,0.000727644,0.0009484519,0.00007745549,0.00006206336],"category_scores_gemma":[0.0006696422,0.0002152695,0.000134543,0.0005109053,0.0002582112,0.001343624,0.0008174733,0.0004979044,0.00006400165],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001087377,"about_ca_system_score_gemma":0.00007398223,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002841452,"about_ca_topic_score_gemma":0.00001775935,"domain_scores_codex":[0.9978467,0.00003270008,0.000661901,0.0006304811,0.00044076,0.0003874932],"domain_scores_gemma":[0.9986459,0.000217458,0.0003053812,0.0002366406,0.0004951085,0.00009948102],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.0001277213,0.0006475234,0.006138423,0.00003900703,0.00004420347,0.00001116149,0.007656305,0.0002327798,0.02261426,0.5893441,0.00006115045,0.3730834],"study_design_scores_gemma":[0.00005467905,0.000659675,0.0009306919,0.0006815679,0.000008140591,0.00002632155,0.01293116,0.06013257,0.8055604,0.1180062,0.0006106467,0.0003979056],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9449553,0.00006545974,0.01335995,0.004848044,0.0005629308,0.0004390805,0.000006072886,0.00007353227,0.03568964],"genre_scores_gemma":[0.9985744,0.00008496622,0.0003078174,0.0002948523,0.00002938022,0.00002565985,0.000001339926,0.00001231928,0.0006692709],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.7829462,"threshold_uncertainty_score":0.8778437,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04371482850026954,"score_gpt":0.2891511786187464,"score_spread":0.2454363501184769,"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."}}