{"id":"W4391845938","doi":"10.1090/bull/1826","title":"Proof in the time of machines","year":2024,"lang":"en","type":"article","venue":"Bulletin of the American Mathematical Society","topic":"Logic, programming, and type systems","field":"Computer Science","cited_by":4,"is_retracted":false,"has_abstract":true,"ca_institutions":"Université de Montréal","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Mathematical proof; Computer-assisted proof; Proof of concept; Proof complexity; Computer science; Analytic proof; Proof assistant; Burden of proof; Proof theory; Structural proof theory; Theoretical computer science; Calculus (dental); Mathematics; Automated theorem proving; Medicine","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":[],"consensus_categories":[],"category_scores_codex":[0.0009243781,0.00009220099,0.0002463053,0.00001154381,0.00003589848,0.0000476515,0.001106689,0.00002363092,0.00005606491],"category_scores_gemma":[0.00009070311,0.00004195891,0.0002730345,0.0004202088,0.0003717363,0.00001589549,0.000226015,0.0001321352,0.00007682298],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00001368443,"about_ca_system_score_gemma":0.00002716817,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00008358007,"about_ca_topic_score_gemma":5.125419e-7,"domain_scores_codex":[0.9989322,0.0001453028,0.0002781707,0.0001621197,0.0003243644,0.000157791],"domain_scores_gemma":[0.9988918,0.0004483906,0.0001393619,0.0004765301,0.00002467927,0.00001921376],"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":[0.000004621727,0.0003933929,0.0003564557,0.0008466887,0.0001018744,0.000004493656,0.01542971,0.00001952673,0.0005845157,0.8969109,0.06240229,0.02294551],"study_design_scores_gemma":[0.0003990167,0.0005216038,0.00275449,0.0002741414,0.00008493548,0.0001146061,0.001954577,0.09928198,0.002026124,0.754047,0.1380101,0.0005314556],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1140751,0.002496626,0.6825163,0.1133857,0.0006409709,0.003218238,0.000009859337,0.0004915756,0.08316564],"genre_scores_gemma":[0.9869257,0.000005772524,0.01158026,0.0004886035,0.00003899834,0.00002417994,2.050687e-7,0.00000780603,0.0009284431],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.8728506,"threshold_uncertainty_score":0.2056523,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.0116543700275107,"score_gpt":0.2478743068915517,"score_spread":0.2362199368640409,"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."}}