{"id":"https://openalex.org/W7160899538","doi":"https://doi.org/10.48550/arxiv.2605.09195","title":"The Geometry of Forgetting: Temporal Knowledge Drift as an Independent Axis in LLM Representations","display_name":"The Geometry of Forgetting: Temporal Knowledge Drift as an Independent Axis in LLM Representations","publication_year":2026,"publication_date":"2026-05-09","ids":{"openalex":"https://openalex.org/W7160899538","doi":"https://doi.org/10.48550/arxiv.2605.09195"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2605.09195","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.09195","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2605.09195","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5119181195","display_name":"Rania Elbadry","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Elbadry, Rania","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135957207","display_name":"Ahmed Heakl","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Heakl, Ahmed","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135950792","display_name":"Fan Zhang","orcid":"https://orcid.org/0000-0002-0343-3499"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhang, Fan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135991395","display_name":"Dani Bouch","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Bouch, Dani","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135929498","display_name":"Yuxia Wang","orcid":"https://orcid.org/0000-0001-5856-144X"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Yuxia","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5135921353","display_name":"Preslav Nakov","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Nakov, Preslav","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5135936167","display_name":"Zhuohan Xie","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xie, Zhuohan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":0,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":false,"cited_by_count":0,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":null,"last_page":null},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.3361000120639801,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},"topics":[{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.3361000120639801,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T12026","display_name":"Explainable Artificial Intelligence (XAI)","score":0.18219999969005585,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T11714","display_name":"Multimodal Machine Learning Applications","score":0.06639999896287918,"subfield":{"id":"https://openalex.org/subfields/1707","display_name":"Computer Vision and Pattern Recognition"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/correctness","display_name":"Correctness","score":0.8086000084877014},{"id":"https://openalex.org/keywords/projection","display_name":"Projection (relational algebra)","score":0.6093000173568726},{"id":"https://openalex.org/keywords/residual","display_name":"Residual","score":0.5483999848365784},{"id":"https://openalex.org/keywords/security-token","display_name":"Security token","score":0.490200012922287},{"id":"https://openalex.org/keywords/interval","display_name":"Interval (graph theory)","score":0.4668999910354614},{"id":"https://openalex.org/keywords/constant","display_name":"Constant (computer programming)","score":0.45350000262260437},{"id":"https://openalex.org/keywords/state","display_name":"State (computer science)","score":0.4214000105857849},{"id":"https://openalex.org/keywords/discriminator","display_name":"Discriminator","score":0.3246999979019165}],"concepts":[{"id":"https://openalex.org/C55439883","wikidata":"https://www.wikidata.org/wiki/Q360812","display_name":"Correctness","level":2,"score":0.8086000084877014},{"id":"https://openalex.org/C57493831","wikidata":"https://www.wikidata.org/wiki/Q3134666","display_name":"Projection (relational algebra)","level":2,"score":0.6093000173568726},{"id":"https://openalex.org/C11413529","wikidata":"https://www.wikidata.org/wiki/Q8366","display_name":"Algorithm","level":1,"score":0.5630000233650208},{"id":"https://openalex.org/C155512373","wikidata":"https://www.wikidata.org/wiki/Q287450","display_name":"Residual","level":2,"score":0.5483999848365784},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.538100004196167},{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.490200012922287},{"id":"https://openalex.org/C2778067643","wikidata":"https://www.wikidata.org/wiki/Q166507","display_name":"Interval (graph theory)","level":2,"score":0.4668999910354614},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.46050000190734863},{"id":"https://openalex.org/C2777027219","wikidata":"https://www.wikidata.org/wiki/Q1284190","display_name":"Constant (computer programming)","level":2,"score":0.45350000262260437},{"id":"https://openalex.org/C48103436","wikidata":"https://www.wikidata.org/wiki/Q599031","display_name":"State (computer science)","level":2,"score":0.4214000105857849},{"id":"https://openalex.org/C31972630","wikidata":"https://www.wikidata.org/wiki/Q844240","display_name":"Computer vision","level":1,"score":0.3995000123977661},{"id":"https://openalex.org/C2779803651","wikidata":"https://www.wikidata.org/wiki/Q5282088","display_name":"Discriminator","level":3,"score":0.3246999979019165},{"id":"https://openalex.org/C153180895","wikidata":"https://www.wikidata.org/wiki/Q7148389","display_name":"Pattern recognition (psychology)","level":2,"score":0.32010000944137573},{"id":"https://openalex.org/C2776760102","wikidata":"https://www.wikidata.org/wiki/Q5139990","display_name":"Code (set theory)","level":3,"score":0.3073999881744385},{"id":"https://openalex.org/C140779682","wikidata":"https://www.wikidata.org/wiki/Q210868","display_name":"Sampling (signal processing)","level":3,"score":0.30570000410079956},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.2989000082015991},{"id":"https://openalex.org/C81669768","wikidata":"https://www.wikidata.org/wiki/Q2359161","display_name":"Precision and recall","level":2,"score":0.29840001463890076},{"id":"https://openalex.org/C2779765954","wikidata":"https://www.wikidata.org/wiki/Q5159625","display_name":"Confabulation (neural networks)","level":3,"score":0.27959999442100525},{"id":"https://openalex.org/C159694833","wikidata":"https://www.wikidata.org/wiki/Q2321565","display_name":"Iterative method","level":2,"score":0.2784999907016754},{"id":"https://openalex.org/C97256817","wikidata":"https://www.wikidata.org/wiki/Q1462316","display_name":"Spurious relationship","level":2,"score":0.27070000767707825},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.2680000066757202},{"id":"https://openalex.org/C140479938","wikidata":"https://www.wikidata.org/wiki/Q5254619","display_name":"Iterated function","level":2,"score":0.2678999900817871},{"id":"https://openalex.org/C115961682","wikidata":"https://www.wikidata.org/wiki/Q860623","display_name":"Image (mathematics)","level":2,"score":0.26100000739097595},{"id":"https://openalex.org/C100660578","wikidata":"https://www.wikidata.org/wiki/Q18733","display_name":"Recall","level":2,"score":0.2529999911785126},{"id":"https://openalex.org/C178009071","wikidata":"https://www.wikidata.org/wiki/Q93344","display_name":"Trigonometric functions","level":2,"score":0.2515000104904175}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2605.09195","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.09195","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2605.09195","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.09195","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Large":[0],"language":[1],"models":[2],"confidently":[3],"produce":[4],"outdated":[5],"answers,":[6],"and":[7,50,93,127,141,161,178,200],"no":[8],"existing":[9],"method":[10,53],"can":[11],"detect":[12],"them.":[13,154],"We":[14,67],"show":[15],"this":[16,69],"is":[17,36,60],"not":[18],"an":[19],"engineering":[20],"failure":[21],"but":[22],"a":[23,29,39],"structural":[24],"one:":[25],"temporal":[26],"drift,":[27],"whether":[28],"stored":[30],"fact":[31],"has":[32],"changed":[33],"since":[34],"training,":[35],"encoded":[37],"as":[38],"direction":[40],"in":[41],"the":[42,103,131,164,166,170,175],"residual":[43],"stream":[44],"geometrically":[45],"orthogonal":[46],"to":[47,63],"both":[48],"correctness":[49,56],"uncertainty.":[51],"Any":[52],"operating":[54],"on":[55,79,87,169],"or":[57],"uncertainty":[58],"signals":[59],"therefore":[61],"blind":[62],"drift":[64,80],"by":[65],"construction.":[66],"verify":[68],"across":[70],"six":[71,146],"instruction-tuned":[72],"models.":[73],"A":[74,155],"linear":[75],"probe":[76,167],"trained":[77],"directly":[78],"labels":[81],"achieves":[82],"AUROC":[83],"$0.83$--$0.95$;":[84],"methods":[85],"based":[86],"token":[88],"entropy,":[89,91],"semantic":[90],"CCS,":[92],"SAPLMA":[94],"all":[95],"remain":[96],"near":[97],"chance":[98],"($0.49$--$0.57$).":[99],"Five":[100],"tests":[101],"confirm":[102],"geometric":[104],"orthogonality:":[105],"weight":[106],"cosines":[107],"($|\\cos|":[108],"\\leq":[109,114,120],"0.14$),":[110],"score":[111],"correlations":[112],"($|r|":[113],"0.20$),":[115],"bidirectional":[116],"null-space":[117,123],"projection":[118,124],"($|\u0394|":[119],"0.008$),":[121],"iterative":[122],"with":[125],"$k{=}10$,":[126],"difference-of-means":[128],"dissociation.":[129],"Mechanistically,":[130],"MLP":[132],"retrieval":[133],"circuit":[134],"produces":[135],"identical":[136],"dynamics":[137],"for":[138],"stale":[139],"recall":[140],"confabulation":[142],"($r":[143],">":[144],"0.81$,":[145],"models),":[147],"explaining":[148],"why":[149],"output":[150],"confidence":[151],"cannot":[152],"separate":[153],"cross-cutoff":[156],"experiment":[157],"holds":[158],"inputs":[159],"constant":[160],"varies":[162],"only":[163],"model:":[165],"fires":[168],"model":[171,186],"whose":[172],"training":[173],"predates":[174],"fact's":[176],"transition":[177],"stays":[179],"silent":[180],"otherwise":[181],"($P(A{>}B)":[182],"=":[183],"0.975$--$0.998$,":[184],"twelve":[185],"pairs),":[187],"confirming":[188],"it":[189],"reads":[190],"model-internal":[191],"knowledge":[192],"state":[193],"rather":[194],"than":[195],"input":[196],"properties.":[197],"Our":[198],"code":[199],"datasets":[201],"will":[202],"be":[203],"publicly":[204],"released.":[205]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-05-13T00:00:00"}