{"id":"W6887927589","doi":"10.17863/cam.108012","title":"On the Power of Interactive Proofs for Learning","year":2024,"lang":"en","type":"other","venue":"Apollo (University of Cambridge)","topic":"","field":"","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Engineering and Physical Sciences Research Council; Natural Sciences and Engineering Research Council of Canada; University of Warwick; UK Research and Innovation","keywords":"Mathematical proof; Multiplicative function; Class (philosophy); Protocol (science); Function (biology); Construct (python library); Concept class; Boolean function","routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.0002724119,0.0002840407,0.0005072524,0.0005827536,0.00006943275,0.0000123291,0.0005349067,0.0002239482,0.00151983],"category_scores_gemma":[0.000135276,0.0002664286,0.0004286277,0.0003263929,0.0003293625,0.00007109184,0.0001960684,0.0004883247,0.002473913],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001344522,"about_ca_system_score_gemma":0.0001113482,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0008465065,"about_ca_topic_score_gemma":0.0001488742,"domain_scores_codex":[0.9987434,0.0001168668,0.0001248246,0.0004120127,0.0003571784,0.0002457076],"domain_scores_gemma":[0.998525,0.0002865609,0.0005689057,0.0004441978,0.0001207338,0.00005455194],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.000361017,0.00007685531,0.00001027582,0.000239988,0.000787684,0.00001375129,0.001534848,0.000007223045,0.0004870647,0.01836343,0.9778345,0.0002833565],"study_design_scores_gemma":[0.0007424679,0.0004413431,0.00009205323,0.001452369,0.0004044293,0.000003473511,0.007005672,0.0001607962,0.0003276826,0.0001907686,0.9888324,0.0003465954],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"other","genre_gemma":"other","genre_scores_codex":[0.004730577,0.0003289408,0.0006163534,0.0004205692,0.0003821075,0.001190897,0.001162063,0.0002364623,0.990932],"genre_scores_gemma":[0.1760734,0.00001163155,0.0002304215,0.00002187553,0.00004311806,0.000001169034,0.00004166116,0.0005198563,0.8230569],"genre_candidate":"other","genre_consensus":"other","teacher_disagreement_score":0.1713428,"threshold_uncertainty_score":0.9999788,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.0108469763287367,"score_gpt":0.2246719435668673,"score_spread":0.2138249672381306,"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."}}