{"id":"W4396671403","doi":"10.26434/chemrxiv-2024-h5sw5","title":"Relative Generality and Risk: Quantitative Measures For Broad Catalyst Success","year":2024,"lang":"en","type":"preprint","venue":"ChemRxiv","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia","funders":"Natural Sciences and Engineering Research Council of Canada; University of British Columbia; Canada Foundation for Innovation; Compute Canada","keywords":"Generality; Risk analysis (engineering); Business; Economics; Management","routes":{"ca_aff":true,"ca_fund":true,"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":["metaresearch","metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.001254495,0.0003560446,0.0006729417,0.00007283871,0.0001169337,0.0001372642,0.0002340614,0.0003308043,0.00003401249],"category_scores_gemma":[0.008962198,0.0002906107,0.0001782406,0.0001001953,0.0002530933,0.00003820285,0.0005800727,0.0008416608,0.0000134238],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00006907258,"about_ca_system_score_gemma":0.0001187939,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0003240769,"about_ca_topic_score_gemma":0.00007585374,"domain_scores_codex":[0.9980731,0.0002185623,0.0004783551,0.0007404028,0.0002249496,0.0002646268],"domain_scores_gemma":[0.9951652,0.003634381,0.0003029509,0.0004661746,0.0003056869,0.0001255818],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.00006932546,0.00007058359,0.001157784,0.002921453,0.0006694427,0.000006563692,0.002140826,0.000003988976,0.0003116788,0.9807909,0.003570424,0.008287074],"study_design_scores_gemma":[0.0002004055,0.00004826034,0.0009757345,0.0003607413,0.0005957175,0.00000154022,0.00008885386,0.002639967,0.005379559,0.988989,0.0003532273,0.0003670056],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.3771135,0.00327429,0.6115438,0.0003558334,0.0007835689,0.001170431,0.0008026889,0.0001795077,0.004776407],"genre_scores_gemma":[0.2608844,0.0003872298,0.7372572,0.00002562985,0.0002040119,0.0003831611,0.00005357947,0.00007417486,0.000730647],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.1257134,"threshold_uncertainty_score":0.9999546,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1989502122721,"score_gpt":0.4352064951061423,"score_spread":0.2362562828340423,"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."}}