{"id":"W4412889746","doi":"10.18653/v1/2025.acl-long.1458","title":"Stochastic Chameleons: Irrelevant Context Hallucinations Reveal Class-Based (Mis)Generalization in LLMs","year":2025,"lang":"en","type":"article","venue":"","topic":"Anomaly Detection Techniques and Applications","field":"Computer Science","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Canadian Institute for Advanced Research; Nvidia","keywords":"Generalization; Context (archaeology); Class (philosophy); Computer science; Cognitive psychology; Artificial intelligence; Psychology; Social psychology; Econometrics; Mathematics; Geography","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":[],"consensus_categories":[],"category_scores_codex":[0.0001763519,0.0001124976,0.0001296537,0.0003458748,0.000144814,0.00009089702,0.0004071505,0.00007841348,0.00002936543],"category_scores_gemma":[0.00004700103,0.0001101371,0.000051372,0.001124751,0.00003506967,0.0001937165,0.00005457757,0.0001145295,0.00002848987],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000116536,"about_ca_system_score_gemma":0.000141572,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000155858,"about_ca_topic_score_gemma":0.0003077412,"domain_scores_codex":[0.9989969,0.00004373262,0.0003035757,0.0003460862,0.0001286062,0.0001811111],"domain_scores_gemma":[0.9992319,0.00008500885,0.00006900348,0.0004504325,0.0001197688,0.00004384604],"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":[0.000003114853,0.0001317696,0.000138022,0.00001205887,0.00000542705,9.332486e-7,0.00007116068,0.003425541,0.001424843,0.9553312,0.00260084,0.03685509],"study_design_scores_gemma":[0.0005622939,0.0000677642,0.003484739,0.00006474876,0.000006690262,0.000002242238,0.00006908245,0.9584899,0.0067677,0.0178326,0.01241366,0.0002385706],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.00181127,0.00005659447,0.9850476,0.006766631,0.00009814846,0.0003983639,0.000002437394,0.0004046693,0.005414233],"genre_scores_gemma":[0.9712028,0.000004966184,0.02335844,0.00185828,0.0000139484,0.0002891385,0.000008027317,0.000005622443,0.003258766],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.9693915,"threshold_uncertainty_score":0.4491262,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.009901039363788106,"score_gpt":0.2679487081815004,"score_spread":0.2580476688177122,"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."}}