{"id":"W4403346052","doi":"10.48550/arxiv.2410.01444","title":"Geometric Signatures of Compositionality Across a Language Model's Lifetime","year":2024,"lang":"en","type":"preprint","venue":"arXiv (Cornell University)","topic":"Syntax, Semantics, Linguistic Variation","field":"Arts and Humanities","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Fonds de recherche du Québec – Nature et technologies; European Commission; Universitat Pompeu Fabra; Wellcome Trust","keywords":"Principle of compositionality; Computer science; Natural language processing; Linguistics; Artificial intelligence; Philosophy","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.003490609,0.0003801062,0.0006643349,0.001164832,0.000981718,0.003568889,0.0008834986,0.0007051653,0.002944856],"category_scores_gemma":[0.02139478,0.0004632823,0.0007332863,0.0009687089,0.00302229,0.007487938,0.003514151,0.002304899,0.0005665581],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001396751,"about_ca_system_score_gemma":0.0008510452,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001514475,"about_ca_topic_score_gemma":0.002124188,"domain_scores_codex":[0.9978946,0.0008389481,0.0001274038,0.0006058313,0.0003580935,0.0001752472],"domain_scores_gemma":[0.9890447,0.005390399,0.001132131,0.003280761,0.0007014162,0.0004507145],"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":[0.0007058486,0.0001455523,0.05622625,0.0003324174,0.000289645,0.0006206926,0.005851579,0.08606049,0.02393932,0.6877575,0.004515088,0.1335556],"study_design_scores_gemma":[0.00002571161,0.0001221796,0.01463675,0.00004356824,0.00006268289,0.0004141129,0.0008239354,0.2973956,0.004994672,0.6748105,0.006597627,0.00007262181],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7018343,0.0005571705,0.2816149,0.002343356,0.00005369917,0.00004212658,0.001694756,0.001014442,0.01084518],"genre_scores_gemma":[0.9769145,0.0001978877,0.02020969,0.0001401662,0.00003679773,0.00007865478,0.001059309,0.000212106,0.00115099],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.003568889,"threshold_uncertainty_score":0.01846033,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.06210371445777068,"score_gpt":0.2193220437224882,"score_spread":0.1572183292647175,"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."}}