{"id":"W7106800886","doi":"10.48448/6sdn-j688","title":"SparsePO: Controlling Preference Alignment of LLMs via Sparse Token Masks","year":2025,"lang":"","type":"other","venue":"Open MIND","topic":"","field":"","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Artificial Intelligence in Medicine (Canada)","funders":"","keywords":"Automatic summarization; Security token; Preference; Divergence (linguistics); Sequence (biology); Word (group theory); Sequence labeling; Helpfulness","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001692605,0.001101039,0.0008089484,0.0004675292,0.0004023475,0.001162171,0.001642494,0.001217847,0.005177497],"category_scores_gemma":[0.009414324,0.0004951082,0.000596681,0.0004332805,0.0007998968,0.002629549,0.001947888,0.001622304,0.002075215],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0008540318,"about_ca_system_score_gemma":0.001277774,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002020932,"about_ca_topic_score_gemma":0.004020526,"domain_scores_codex":[0.9988415,0.0004532427,0.00006641437,0.000317901,0.0002040973,0.0001169483],"domain_scores_gemma":[0.9976078,0.001276894,0.0003083807,0.000368136,0.0002685131,0.0001702131],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.001160319,0.0003144931,0.004062939,0.000343893,0.0001260655,0.0002314566,0.0004859799,0.4532245,0.04357782,0.02503065,0.008711094,0.4627309],"study_design_scores_gemma":[0.00002719723,0.00008070026,0.0002176757,0.000008094486,0.000009637168,0.00003021614,0.00002791758,0.9834349,0.005651151,0.009440308,0.001059263,0.00001300207],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.03241049,0.0001648243,0.961181,0.0002582838,0.0000591744,0.0000894535,0.0001562682,0.003791092,0.001889348],"genre_scores_gemma":[0.740666,0.0001096826,0.2509085,0.0004000505,0.00008508052,0.0002748069,0.0005220082,0.000887537,0.00614622],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.005177497,"threshold_uncertainty_score":0.01732045,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.08704859168442897,"score_gpt":0.3149941125704026,"score_spread":0.2279455208859736,"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."}}