{"id":"W4393161644","doi":"10.1609/aaai.v38i21.30437","title":"A SAT + Computer Algebra System Verification of the Ramsey Problem R(3, 8) (Student Abstract)","year":2024,"lang":"en","type":"article","venue":"Proceedings of the AAAI Conference on Artificial Intelligence","topic":"Computability, Logic, AI Algorithms","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Windsor; University of Waterloo","funders":"","keywords":"Computer science; Algebra over a field; Symbolic computation; Programming language; Calculus (dental); Theoretical computer science; Mathematics; Pure mathematics","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":[0.001086548,0.0003079583,0.0003686332,0.0001403596,0.0001992515,0.0005656182,0.003902083,0.0001272156,0.00002358738],"category_scores_gemma":[0.00009127081,0.0001932137,0.0002725377,0.001162977,0.0003897106,0.0004323023,0.001076026,0.0004561652,0.0001023585],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001736137,"about_ca_system_score_gemma":0.0002152515,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00006606516,"about_ca_topic_score_gemma":0.000008832913,"domain_scores_codex":[0.9968417,0.00004265185,0.0009897085,0.0007960658,0.0009550871,0.0003747592],"domain_scores_gemma":[0.9977249,0.0002012592,0.0004957141,0.0007068176,0.0007870129,0.00008432965],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00001560893,0.0002024285,0.0002544861,0.0004263427,0.00003614084,7.565435e-7,0.002993293,0.000303345,0.01446359,0.8814842,0.0001559124,0.09966396],"study_design_scores_gemma":[0.00005014752,0.000326224,0.007219065,0.001973883,0.00005056716,0.00003116174,0.000889777,0.5423726,0.325164,0.1211264,0.000311478,0.0004847122],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6736377,0.0003190956,0.2868593,0.01225185,0.006834319,0.004066685,0.00003024725,0.001024503,0.01497632],"genre_scores_gemma":[0.9928177,0.00001173405,0.006817637,0.00006954847,0.0001458889,0.00004640263,4.824656e-7,0.00001679368,0.00007380078],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.7603578,"threshold_uncertainty_score":0.7879027,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04920799804165031,"score_gpt":0.2839430372953485,"score_spread":0.2347350392536982,"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."}}