https://chiosa.app/papers/ask-about-the-right-moment-training-free-object-identity-reasoning-for-video-language-models-2610-31471Chiosaen2026-10-23T15:58:10ZA tracker helps video models keep object identities through a shufflehttps://chiosa.app/papers/ask-about-the-right-moment-training-free-object-identity-reasoning-for-video-language-models-2610-31471?lang=esChiosaes2026-10-23T15:58:10ZUn rastreador ayuda a modelos de video a conservar la identidad de los objetos al barajarhttps://chiosa.app/papers/ask-about-the-right-moment-training-free-object-identity-reasoning-for-video-language-models-2610-31471?lang=frChiosafr2026-10-23T15:58:10ZUn système de suivi aide les modèles vidéo à conserver l’identité des objets pendant un mélangehttps://chiosa.app/papers/compiling-autoregressive-llm-control-into-policy-exact-parallel-inference-for-hierarchical-rag-2610-30664Chiosaen2026-10-23T15:58:10ZParallel compilation speeds finite-action language model controlhttps://chiosa.app/papers/drill-training-world-models-to-improve-policies-not-predict-pixels-2610-30814Chiosaen2026-10-23T15:58:10ZTraining video world models for robot learning, not just clearer frameshttps://chiosa.app/papers/energy-and-latency-first-how-deployment-constraints-reverse-time-series-model-selection-2610-30965Chiosaen2026-10-23T15:58:10ZDeadline and energy reshape time-series model choicehttps://chiosa.app/papers/focusvla-future-intent-guided-dynamic-view-routing-for-vision-language-action-models-2610-31423Chiosaen2026-10-23T15:58:10ZA robot method routes camera views by future action needshttps://chiosa.app/papers/focusvla-future-intent-guided-dynamic-view-routing-for-vision-language-action-models-2610-31423?lang=esChiosaes2026-10-23T15:58:10ZUn método adapta las vistas de cámara a las acciones del robothttps://chiosa.app/papers/random-remasking-scales-masked-diffusion-inference-2610-30744Chiosaen2026-10-23T15:58:10ZRandom remasking lets diffusion models revise and improve sequenceshttps://chiosa.app/papers/rift-kv-cache-compression-for-draft-models-with-complementary-compressibility-training-2610-31098Chiosaen2026-10-23T15:58:10ZRIFT keeps draft caches useful as generation growshttps://chiosa.app/papers/rift-kv-cache-compression-for-draft-models-with-complementary-compressibility-training-2610-31098?lang=itChiosait2026-10-23T15:58:10ZRIFT mantiene utile la cache della bozza mentre il testo crescehttps://chiosa.app/papers/rift-kv-cache-compression-for-draft-models-with-complementary-compressibility-training-2610-31098?lang=esChiosaes2026-10-23T15:58:10ZRIFT mantiene útil la caché del borrador al generarhttps://chiosa.app/papers/ripple-recovery-internalization-through-process-level-policy-learning-and-effect-aware-credit-for-long-horizo-2610-31276Chiosaen2026-10-23T15:58:10ZCoding agents can learn to turn failure critiques into repairshttps://chiosa.app/papers/ripple-recovery-internalization-through-process-level-policy-learning-and-effect-aware-credit-for-long-horizo-2610-31276?lang=itChiosait2026-10-23T15:58:10ZGli agenti di programmazione possono imparare a trasformare le critiche agli errori in riparazionihttps://chiosa.app/papers/ripple-recovery-internalization-through-process-level-policy-learning-and-effect-aware-credit-for-long-horizo-2610-31276?lang=esChiosaes2026-10-23T15:58:10ZLos agentes de programación pueden aprender a convertir críticas de fallos en reparacioneshttps://chiosa.app/papers/same-entity-different-script-diagnosing-cross-script-entity-matching-failures-in-language-models-2610-31407Chiosaen2026-10-23T15:58:10ZIdentity matching weakens when names switch scriptshttps://chiosa.app/papers/same-entity-different-script-diagnosing-cross-script-entity-matching-failures-in-language-models-2610-31407?lang=itChiosait2026-10-23T15:58:10ZIl riconoscimento delle identità cala quando cambiano le scritture dei nomihttps://chiosa.app/papers/same-entity-different-script-diagnosing-cross-script-entity-matching-failures-in-language-models-2610-31407?lang=esChiosaes2026-10-23T15:58:10ZLa coincidencia de identidades falla cuando cambian los sistemas de escriturahttps://chiosa.app/papers/same-entity-different-script-diagnosing-cross-script-entity-matching-failures-in-language-models-2610-31407?lang=frChiosafr2026-10-23T15:58:10ZLa correspondance d’identité faiblit quand l’écriture des noms changehttps://chiosa.app/papers/spectralcanary-embarrassingly-simple-low-frequency-traces-for-t2i-dataset-use-auditing-2610-31659Chiosaen2026-10-23T15:58:10ZA shared image trace helps reveal when models learn a collection's stylehttps://chiosa.app/papers/spectralcanary-embarrassingly-simple-low-frequency-traces-for-t2i-dataset-use-auditing-2610-31659?lang=esChiosaes2026-10-23T15:58:10ZUna señal compartida ayuda a revelar cuándo los modelos aprenden el estilo de una colecciónhttps://chiosa.app/papers/spectralcanary-embarrassingly-simple-low-frequency-traces-for-t2i-dataset-use-auditing-2610-31659?lang=frChiosafr2026-10-23T15:58:10ZUn signal visuel commun aide à repérer quand un modèle apprend le style d’une collectionhttps://chiosa.app/papers/typed-template-fuzzing-2610-31342Chiosaen2026-10-23T15:58:10ZFuzzing from typed templates reaches many valid programshttps://chiosa.app/papers/typed-template-fuzzing-2610-31342?lang=frChiosafr2026-10-23T15:58:10ZLe fuzzing par modèles typés produit de nombreux programmes valideshttps://chiosa.app/papers/uniguard-multi-view-guardrail-learning-for-cross-interface-safety-generalization-2610-30742Chiosaen2026-10-23T15:58:10ZTraining guardrails across formats improves safety judgmentshttps://chiosa.app/papers/when-is-a-memory-safe-enough-to-act-on-action-conditioned-verification-for-long-lived-agents-2610-31446Chiosaen2026-10-23T15:58:10ZA gate checks memories before consequential actionshttps://chiosa.app/papers/when-is-a-memory-safe-enough-to-act-on-action-conditioned-verification-for-long-lived-agents-2610-31446?lang=esChiosaes2026-10-23T15:58:10ZUna compuerta comprueba los recuerdos antes de acciones importanteshttps://chiosa.app/papers/when-is-a-memory-safe-enough-to-act-on-action-conditioned-verification-for-long-lived-agents-2610-31446?lang=frChiosafr2026-10-23T15:58:10ZUn filtre vérifie la mémoire avant une action lourde de conséquenceshttps://chiosa.app/papers/afterdeletebench-auditing-post-deletion-information-persistence-in-llm-agents-2610-31366Chiosaen2026-10-23T10:33:44ZDeleting a record can leave its secret recoverable by an agenthttps://chiosa.app/papers/afterdeletebench-auditing-post-deletion-information-persistence-in-llm-agents-2610-31366?lang=itChiosait2026-10-23T10:33:44ZEliminare un documento può lasciare il segreto recuperabile da un agentehttps://chiosa.app/papers/afterdeletebench-auditing-post-deletion-information-persistence-in-llm-agents-2610-31366?lang=esChiosaes2026-10-23T10:33:44ZBorrar un registro puede dejar su secreto al alcance de un agentehttps://chiosa.app/papers/afterdeletebench-auditing-post-deletion-information-persistence-in-llm-agents-2610-31366?lang=frChiosafr2026-10-23T10:33:44ZSupprimer un enregistrement peut laisser un secret récupérable par un agenthttps://chiosa.app/papers/agenttriage-replayable-security-diagnosis-across-heterogeneous-agent-runtimes-2610-30738Chiosaen2026-10-23T10:33:44ZAgentTriage makes agent security diagnoses replayablehttps://chiosa.app/papers/auditing-the-hand-written-inputs-of-learned-access-control-2610-31388Chiosaen2026-10-23T10:33:44ZAudits expose risks in hand-written access controlhttps://chiosa.app/papers/auditing-the-hand-written-inputs-of-learned-access-control-2610-31388?lang=itChiosait2026-10-23T10:33:44ZGli audit rivelano rischi nelle regole manuali per il controllo degli accessihttps://chiosa.app/papers/auditing-the-hand-written-inputs-of-learned-access-control-2610-31388?lang=esChiosaes2026-10-23T10:33:44ZAuditorías revelan riesgos en el control de acceso manualhttps://chiosa.app/papers/auditing-the-hand-written-inputs-of-learned-access-control-2610-31388?lang=frChiosafr2026-10-23T10:33:44ZDes audits révèlent les risques des règles écrites à la main en contrôle d’accèshttps://chiosa.app/papers/automating-feature-flag-cleanup-at-scale-with-multi-agent-llm-systems-an-industrial-experience-at-doordash-2610-31351Chiosaen2026-10-23T10:33:44ZDoorDash system automates feature flag cleanup across codehttps://chiosa.app/papers/automating-feature-flag-cleanup-at-scale-with-multi-agent-llm-systems-an-industrial-experience-at-doordash-2610-31351?lang=itChiosait2026-10-23T10:33:44ZDoorDash automatizza la rimozione dei feature flag tra più filehttps://chiosa.app/papers/automating-feature-flag-cleanup-at-scale-with-multi-agent-llm-systems-an-industrial-experience-at-doordash-2610-31351?lang=esChiosaes2026-10-23T10:33:44ZDoorDash automatiza la limpieza de indicadores en el códigohttps://chiosa.app/papers/automating-feature-flag-cleanup-at-scale-with-multi-agent-llm-systems-an-industrial-experience-at-doordash-2610-31351?lang=frChiosafr2026-10-23T10:33:44ZDoorDash automatise le nettoyage des indicateurs dans le codehttps://chiosa.app/papers/dispersive-tree-agreements-for-multi-robot-coordination-2610-31540Chiosaen2026-10-23T10:33:44ZRobots can negotiate near-optimal plans while moving in real time (F1)https://chiosa.app/papers/dispersive-tree-agreements-for-multi-robot-coordination-2610-31540?lang=itChiosait2026-10-23T10:33:44ZI robot possono negoziare piani quasi ottimali in tempo reale (F1)https://chiosa.app/papers/dispersive-tree-agreements-for-multi-robot-coordination-2610-31540?lang=esChiosaes2026-10-23T10:33:44ZLos robots pueden negociar planes casi óptimos mientras se mueven en tiempo real (F1)https://chiosa.app/papers/evaluating-the-applicability-of-llm-inference-optimizations-on-apple-silicon-2610-31479Chiosaen2026-10-23T10:33:44ZApple Silicon speedups depend on the workload and runtimehttps://chiosa.app/papers/extreme-compression-of-per-layer-embeddings-via-cross-layer-latent-vector-quantization-2610-31309Chiosaen2026-10-23T10:33:44ZShared token codes sharply compress Gemma 4 embeddingshttps://chiosa.app/papers/extreme-compression-of-per-layer-embeddings-via-cross-layer-latent-vector-quantization-2610-31309?lang=itChiosait2026-10-23T10:33:44ZCodici condivisi comprimono le embedding di Gemma 4https://chiosa.app/papers/extreme-compression-of-per-layer-embeddings-via-cross-layer-latent-vector-quantization-2610-31309?lang=esChiosaes2026-10-23T10:33:44ZCódigos compartidos comprimen las embeddings de Gemma 4https://chiosa.app/papers/extreme-compression-of-per-layer-embeddings-via-cross-layer-latent-vector-quantization-2610-31309?lang=frChiosafr2026-10-23T10:33:44ZDes codes partagés compressent fortement les embeddings de Gemma 4https://chiosa.app/papers/foundation4d-towards-a-unified-perception-and-reconstruction-model-for-autonomous-vehicles-2610-30340Chiosaen2026-10-23T10:33:44ZOne model links driving-scene understanding with 3D reconstructionhttps://chiosa.app/papers/invisibilization-as-a-source-of-probability-bias-in-language-models-2610-30785Chiosaen2026-10-23T10:33:44ZA benchmark reveals when language models set reports asidehttps://chiosa.app/papers/joint-or-per-option-compact-encoders-for-closed-set-decisions-from-17-million-to-1-billion-parameters-2610-30970Chiosaen2026-10-23T10:33:44ZCompact encoders face a trade-off in how they read optionshttps://chiosa.app/papers/joint-or-per-option-compact-encoders-for-closed-set-decisions-from-17-million-to-1-billion-parameters-2610-30970?lang=esChiosaes2026-10-23T10:33:44ZLos codificadores compactos afrontan un dilema al leer opcioneshttps://chiosa.app/papers/large-language-models-show-systematic-failures-in-incorporating-evolving-medical-knowledge-2610-30631Chiosaen2026-10-23T10:33:44ZModels can spot changed guidelines without fully updating advicehttps://chiosa.app/papers/metacognitive-scaffolds-create-productive-friction-for-ai-assisted-knowledge-work-2610-30945Chiosaen2026-10-23T10:33:44ZAn editable writing scaffold improved brief quality and later promptshttps://chiosa.app/papers/not-where-you-think-protecting-location-privacy-against-reasoning-based-geolocation-2610-30439Chiosaen2026-10-23T10:33:44ZModels can locate ordinary photos, and edits can mislead themhttps://chiosa.app/papers/on-the-token-collapse-phenomenon-in-textual-graph-soft-prompting-2610-31572Chiosaen2026-10-23T10:33:44ZGraph prompts can collapse into nearly identical tokenshttps://chiosa.app/papers/openattention-one-percent-is-enough-with-tail-proxies-2610-30951Chiosaen2026-10-23T10:33:44ZMean proxies help sparse attention retain accuracyhttps://chiosa.app/papers/opentac-an-open-ecosystem-for-customizable-scalable-and-transferable-3d-tactile-sensing-2610-31661Chiosaen2026-10-23T10:33:44ZOpenTac helps fit tactile skins to varied robot shapeshttps://chiosa.app/papers/opentac-an-open-ecosystem-for-customizable-scalable-and-transferable-3d-tactile-sensing-2610-31661?lang=itChiosait2026-10-23T10:33:44ZOpenTac adatta le skin tattili a forme diverse di robothttps://chiosa.app/papers/opentac-an-open-ecosystem-for-customizable-scalable-and-transferable-3d-tactile-sensing-2610-31661?lang=frChiosafr2026-10-23T10:33:44ZOpenTac adapte les peaux tactiles aux formes des robotshttps://chiosa.app/papers/practical-auditing-of-differentially-private-graph-release-mechanisms-methodology-toolkit-and-case-study-2610-31294Chiosaen2026-10-23T10:33:44ZAudits found privacy violations in the tested PrivDPR codehttps://chiosa.app/papers/practical-auditing-of-differentially-private-graph-release-mechanisms-methodology-toolkit-and-case-study-2610-31294?lang=itChiosait2026-10-23T10:33:44ZGli audit hanno rilevato violazioni di privacy nel codice testato di PrivDPRhttps://chiosa.app/papers/practical-auditing-of-differentially-private-graph-release-mechanisms-methodology-toolkit-and-case-study-2610-31294?lang=esChiosaes2026-10-23T10:33:44ZLa auditoría detectó fallos de privacidad en el código probado de PrivDPRhttps://chiosa.app/papers/practical-auditing-of-differentially-private-graph-release-mechanisms-methodology-toolkit-and-case-study-2610-31294?lang=frChiosafr2026-10-23T10:33:44ZL’audit révèle des violations de confidentialité dans le code PrivDPR testéhttps://chiosa.app/papers/productwebbench-evaluating-website-continuity-beyond-task-completion-2610-31118Chiosaen2026-10-23T10:33:44ZFrontend changes can pass while website behavior breakshttps://chiosa.app/papers/productwebbench-evaluating-website-continuity-beyond-task-completion-2610-31118?lang=itChiosait2026-10-23T10:33:44ZLe modifiche frontend possono superare i controlli mentre il sito si rompehttps://chiosa.app/papers/productwebbench-evaluating-website-continuity-beyond-task-completion-2610-31118?lang=esChiosaes2026-10-23T10:33:44ZLos cambios frontend pueden pasar mientras se rompe el sitiohttps://chiosa.app/papers/repair-changes-which-layers-to-remove-exact-optima-for-depth-pruning-of-language-models-2610-30552Chiosaen2026-10-23T10:33:44ZLayer repair changes which transformer layers are best to removehttps://chiosa.app/papers/same-luck-different-choice-twin-world-credit-assignment-for-agentic-reinforcement-learning-2610-31544Chiosaen2026-10-23T10:33:44ZTwice gives search agents credit for individual turnshttps://chiosa.app/papers/same-luck-different-choice-twin-world-credit-assignment-for-agentic-reinforcement-learning-2610-31544?lang=itChiosait2026-10-23T10:33:44ZTwice attribuisce agli agenti di ricerca il merito dei singoli turnihttps://chiosa.app/papers/same-luck-different-choice-twin-world-credit-assignment-for-agentic-reinforcement-learning-2610-31544?lang=esChiosaes2026-10-23T10:33:44ZTwice asigna crédito a cada turno de los agentes de búsquedahttps://chiosa.app/papers/same-luck-different-choice-twin-world-credit-assignment-for-agentic-reinforcement-learning-2610-31544?lang=frChiosafr2026-10-23T10:33:44ZTwice attribue aux agents de recherche le mérite de chaque tourhttps://chiosa.app/papers/scaling-laws-for-safety-probes-2610-30678Chiosaen2026-10-23T10:33:44ZScaling laws estimate safety-probe error from size and labelshttps://chiosa.app/papers/simulated-control-evals-2610-31593Chiosaen2026-10-23T10:33:44ZSimulated tests predict safeguards on deployment taskshttps://chiosa.app/papers/simulated-control-evals-2610-31593?lang=esChiosaes2026-10-23T10:33:44ZPruebas simuladas predicen salvaguardas en tareas realeshttps://chiosa.app/papers/simulated-control-evals-2610-31593?lang=frChiosafr2026-10-23T10:33:44ZDes tests simulés évaluent les garde-fous en déploiementhttps://chiosa.app/papers/the-delegation-divide-an-open-index-of-ai-agent-delegability-across-occupations-and-countries-2610-30752Chiosaen2026-10-23T10:33:44ZAn index maps which work AI agents could take over worldwidehttps://chiosa.app/papers/the-value-of-system-1-models-in-medicine-2610-31315Chiosaen2026-10-23T10:33:44ZClinical model scores vary by decision typehttps://chiosa.app/papers/the-value-of-system-1-models-in-medicine-2610-31315?lang=esChiosaes2026-10-23T10:33:44ZLas puntuaciones clínicas varían según el tipo de decisiónhttps://chiosa.app/papers/the-value-of-system-1-models-in-medicine-2610-31315?lang=frChiosafr2026-10-23T10:33:44ZLes scores des modèles cliniques varient selon le type de décisionhttps://chiosa.app/papers/trace2world-reconstructing-app-worlds-from-real-interaction-traces-for-scalable-gui-agent-training-2610-30792Chiosaen2026-10-23T10:33:44ZTrace2World builds browser app worlds from interaction traceshttps://chiosa.app/papers/ulme-predictive-coupling-for-mutually-reinforcing-embedding-and-language-modeling-2610-30705Chiosaen2026-10-23T10:33:44ZOne language model learns retrieval and answer generationhttps://chiosa.app/papers/alps-visual-state-encoders-for-world-models-2610-30120Chiosaen2026-10-22T18:08:13ZOne visual encoder links scene state, language, and image detailhttps://chiosa.app/papers/did-the-problem-actually-get-harder-testing-llms-on-paired-p-and-np-complete-problems-hardened-by-adversarial-2610-30103Chiosaen2026-10-22T18:08:13ZModels solve fewer problems after a structural change makes them NP-completehttps://chiosa.app/papers/did-the-problem-actually-get-harder-testing-llms-on-paired-p-and-np-complete-problems-hardened-by-adversarial-2610-30103?lang=esChiosaes2026-10-22T18:08:13ZLos modelos resuelven menos problemas cuando un cambio estructural los vuelve NP-completoshttps://chiosa.app/papers/donic-generalist-dexterous-bimanual-tracking-control-from-human-demonstrations-2610-29997Chiosaen2026-10-22T18:08:13ZOne robot policy tracks two-hand object referenceshttps://chiosa.app/papers/donic-generalist-dexterous-bimanual-tracking-control-from-human-demonstrations-2610-29997?lang=itChiosait2026-10-22T18:08:13ZUna policy robotica segue riferimenti a oggetti con due manihttps://chiosa.app/papers/donic-generalist-dexterous-bimanual-tracking-control-from-human-demonstrations-2610-29997?lang=esChiosaes2026-10-22T18:08:13ZUna política robótica sigue referencias de objetos con dos manoshttps://chiosa.app/papers/donic-generalist-dexterous-bimanual-tracking-control-from-human-demonstrations-2610-29997?lang=frChiosafr2026-10-22T18:08:13ZUn contrôleur robotique suit des références à deux mainshttps://chiosa.app/papers/dynaweave-learning-to-adapt-through-dynamic-scenario-synthesis-2610-29443Chiosaen2026-10-22T18:08:13ZDynaWeave turns static tool tasks into changing scenarioshttps://chiosa.app/papers/dynaweave-learning-to-adapt-through-dynamic-scenario-synthesis-2610-29443?lang=esChiosaes2026-10-22T18:08:13ZDynaWeave convierte tareas estáticas con herramientas en escenarios cambianteshttps://chiosa.app/papers/encoded-but-unused-bounding-the-sources-of-vlm-errors-2610-29241Chiosaen2026-10-22T18:08:13ZMany tested object-attribute errors show detectable visual evidencehttps://chiosa.app/papers/environment-duels-a-microcosm-of-llm-co-evolution-2610-30116Chiosaen2026-10-22T18:08:13ZModels that solve best are not always the best task authorshttps://chiosa.app/papers/environment-duels-a-microcosm-of-llm-co-evolution-2610-30116?lang=itChiosait2026-10-22T18:08:13ZI modelli che risolvono meglio non sono sempre i migliori autori di compitihttps://chiosa.app/papers/asymmetric-on-policy-distillation-learns-in-context-exploration-2610-30105Chiosaen2026-10-22T15:13:17ZASTEROID teaches partial-view agents to explore from experiencehttps://chiosa.app/papers/asymmetric-on-policy-distillation-learns-in-context-exploration-2610-30105?lang=esChiosaes2026-10-22T15:13:17ZASTEROID enseña a agentes con visión parcial a explorar a partir de la experienciahttps://chiosa.app/papers/asymmetric-on-policy-distillation-learns-in-context-exploration-2610-30105?lang=frChiosafr2026-10-22T15:13:17ZASTEROID apprend aux agents à vue partielle à explorer par l’expériencehttps://chiosa.app/papers/auditing-conformal-prediction-for-language-models-under-inference-time-model-shift-2610-28937Chiosaen2026-10-22T15:13:17ZChanging model settings can reduce prediction coveragehttps://chiosa.app/papers/automating-ml-for-science-can-frontier-agents-climb-scientific-hills-in-the-wild-2610-28763Chiosaen2026-10-22T15:13:17ZA testbed measures coding-agent gains on science modelshttps://chiosa.app/papers/behavior-beyond-text-evaluating-refusal-transfer-failure-in-llm-agents-2610-28982Chiosaen2026-10-22T15:13:17ZChat refusals often fail to prevent unsafe tool callshttps://chiosa.app/papers/better-policies-worse-verifier-audits-2610-29029Chiosaen2026-10-22T15:13:17ZWhen whole-task audits misjudge component checkershttps://chiosa.app/papers/bridge-an-acceptance-gate-to-block-false-improvements-in-model-updates-with-a-retrospective-clinical-evaluati-2610-29515Chiosaen2026-10-22T15:13:17ZA gate checks whether model updates really improvehttps://chiosa.app/papers/bridge-an-acceptance-gate-to-block-false-improvements-in-model-updates-with-a-retrospective-clinical-evaluati-2610-29515?lang=itChiosait2026-10-22T15:13:17ZUn criterio verifica se gli aggiornamenti migliorano davverohttps://chiosa.app/papers/bridge-an-acceptance-gate-to-block-false-improvements-in-model-updates-with-a-retrospective-clinical-evaluati-2610-29515?lang=esChiosaes2026-10-22T15:13:17ZUn filtro comprueba si las actualizaciones de modelos mejoranhttps://chiosa.app/papers/bridge-an-acceptance-gate-to-block-false-improvements-in-model-updates-with-a-retrospective-clinical-evaluati-2610-29515?lang=frChiosafr2026-10-22T15:13:17ZUn filtre vérifie si les mises à jour améliorent vraiment les modèleshttps://chiosa.app/papers/certmoe-robustness-certification-for-mixture-of-experts-llms-via-routing-aware-smoothing-2610-29395Chiosaen2026-10-22T15:13:17ZCertifying robustness when language model routes changehttps://chiosa.app/papers/certmoe-robustness-certification-for-mixture-of-experts-llms-via-routing-aware-smoothing-2610-29395?lang=itChiosait2026-10-22T15:13:17ZCertificare la robustezza quando cambia il routing del modellohttps://chiosa.app/papers/certmoe-robustness-certification-for-mixture-of-experts-llms-via-routing-aware-smoothing-2610-29395?lang=esChiosaes2026-10-22T15:13:17ZCertificar robustez cuando cambian las rutas del modelohttps://chiosa.app/papers/characterizing-and-defending-utility-hijacking-in-instruction-tuning-data-selection-2610-29886Chiosaen2026-10-22T15:13:17ZHow harmful records can evade data selectorshttps://chiosa.app/papers/characterizing-and-defending-utility-hijacking-in-instruction-tuning-data-selection-2610-29886?lang=itChiosait2026-10-22T15:13:17ZCome i dati dannosi possono eludere i selettorihttps://chiosa.app/papers/characterizing-and-defending-utility-hijacking-in-instruction-tuning-data-selection-2610-29886?lang=esChiosaes2026-10-22T15:13:17ZCómo evaden los registros dañinos los selectores de datoshttps://chiosa.app/papers/characterizing-and-defending-utility-hijacking-in-instruction-tuning-data-selection-2610-29886?lang=frChiosafr2026-10-22T15:13:17ZComment des exemples nuisibles peuvent contourner les sélecteurshttps://chiosa.app/papers/compaction-laundering-provenance-destruction-in-context-compacting-coding-agents-2610-29869Chiosaen2026-10-22T15:13:17ZForged labels can turn tool text into user instructionshttps://chiosa.app/papers/compaction-laundering-provenance-destruction-in-context-compacting-coding-agents-2610-29869?lang=esChiosaes2026-10-22T15:13:17ZLas etiquetas falsificadas pueden convertir texto de herramientas en instrucciones del usuariohttps://chiosa.app/papers/constrained-tagging-for-span-labeling-2610-28922Chiosaen2026-10-22T15:13:17ZInline tags help language models locate text spanshttps://chiosa.app/papers/deciding-scale-placement-in-mxfp4-dot-products-with-integer-arithmetic-2610-29445Chiosaen2026-10-22T15:13:17ZWhen compilers can safely move MXFP4 scaleshttps://chiosa.app/papers/deciding-scale-placement-in-mxfp4-dot-products-with-integer-arithmetic-2610-29445?lang=itChiosait2026-10-22T15:13:17ZQuando i compilatori possono spostare in sicurezza le scale MXFP4https://chiosa.app/papers/deciding-scale-placement-in-mxfp4-dot-products-with-integer-arithmetic-2610-29445?lang=esChiosaes2026-10-22T15:13:17ZCuándo puede un compilador mover escalas MXFP4https://chiosa.app/papers/deciding-scale-placement-in-mxfp4-dot-products-with-integer-arithmetic-2610-29445?lang=frChiosafr2026-10-22T15:13:17ZQuand un compilateur peut déplacer les échelles MXFP4 sans risquehttps://chiosa.app/papers/echo-emergent-common-mode-hidden-offsets-drive-vocabulary-collapse-in-preference-optimization-2610-29410Chiosaen2026-10-22T15:13:17ZWhy preference ranking can improve as text generation collapseshttps://chiosa.app/papers/echo-emergent-common-mode-hidden-offsets-drive-vocabulary-collapse-in-preference-optimization-2610-29410?lang=itChiosait2026-10-22T15:13:17ZPerché il ranking delle preferenze può migliorare mentre la generazione collassahttps://chiosa.app/papers/echo-emergent-common-mode-hidden-offsets-drive-vocabulary-collapse-in-preference-optimization-2610-29410?lang=esChiosaes2026-10-22T15:13:17ZPor qué mejora la clasificación por preferencias mientras colapsa la generación de textohttps://chiosa.app/papers/events-not-pixels-where-world-models-get-the-return-wrong-2610-29435Chiosaen2026-10-22T15:13:17ZWorld models can miss events in frames that look accuratehttps://chiosa.app/papers/events-not-pixels-where-world-models-get-the-return-wrong-2610-29435?lang=esChiosaes2026-10-22T15:13:17ZLos modelos del mundo pueden pasar por alto eventos en fotogramas que parecen precisoshttps://chiosa.app/papers/events-not-pixels-where-world-models-get-the-return-wrong-2610-29435?lang=frChiosafr2026-10-22T15:13:17ZLes modèles du monde peuvent manquer des événements dans des images pourtant fidèleshttps://chiosa.app/papers/frontiercs-2-a-benchmark-for-computer-science-r-d-2610-29932Chiosaen2026-10-22T15:13:17ZAgents can improve research code, but many tasks stump allhttps://chiosa.app/papers/frontiercs-2-a-benchmark-for-computer-science-r-d-2610-29932?lang=frChiosafr2026-10-22T15:13:17ZLes agents peuvent améliorer le code de recherche, mais de nombreuses tâches ont laissé tous les agents sans solutionhttps://chiosa.app/papers/huayu-global-skillful-real-time-quasi-global-precipitation-retrieval-2610-29213Chiosaen2026-10-22T15:13:17ZA satellite system brings stronger rainfall detection into real timehttps://chiosa.app/papers/is-2-4-sparsity-still-worth-it-semi-structured-sparsity-versus-quantization-in-llm-inference-2610-28947Chiosaen2026-10-22T15:13:17ZFor LLM inference, lower precision beats 2:4 above the floorhttps://chiosa.app/papers/pause-search-free-steering-of-coding-agents-before-their-first-source-code-edit-2610-29870Chiosaen2026-10-22T15:13:17ZA single repair plan helps coding agents resolve more taskshttps://chiosa.app/papers/pause-search-free-steering-of-coding-agents-before-their-first-source-code-edit-2610-29870?lang=itChiosait2026-10-22T15:13:17ZUn piano di riparazione aiuta gli agenti di coding a risolvere più attivitàhttps://chiosa.app/papers/pause-search-free-steering-of-coding-agents-before-their-first-source-code-edit-2610-29870?lang=esChiosaes2026-10-22T15:13:17ZUn plan de reparación ayuda a los agentes de programación a resolver más tareashttps://chiosa.app/papers/rubatocodec-budgeted-adaptive-speech-tokenization-with-streaming-decoding-2610-29434Chiosaen2026-10-22T15:13:17ZA fixed event budget lets speech codes adapt their timinghttps://chiosa.app/papers/rubatocodec-budgeted-adaptive-speech-tokenization-with-streaming-decoding-2610-29434?lang=itChiosait2026-10-22T15:13:17ZUn event budget fisso adatta i tempi dei codici vocalihttps://chiosa.app/papers/rubatocodec-budgeted-adaptive-speech-tokenization-with-streaming-decoding-2610-29434?lang=esChiosaes2026-10-22T15:13:17ZUn presupuesto fijo de eventos permite adaptar la temporización de los códigos de hablahttps://chiosa.app/papers/rubatocodec-budgeted-adaptive-speech-tokenization-with-streaming-decoding-2610-29434?lang=frChiosafr2026-10-22T15:13:17ZUn budget d’événements fixe permet aux codes vocaux d’adapter leur rythmehttps://chiosa.app/papers/safter-safety-aware-filtering-through-explicit-recoverability-for-robust-reinforcement-learning-under-distrib-2610-29149Chiosaen2026-10-22T15:13:17ZA robot safety filter learns to preserve recovery optionshttps://chiosa.app/papers/tenetbench-a-clause-anchored-russian-corpus-for-evaluating-language-models-against-professional-codes-of-ethi-2610-29477Chiosaen2026-10-22T15:13:17ZA clause-based test exposes conflicts with professional ethics ruleshttps://chiosa.app/papers/tenetbench-a-clause-anchored-russian-corpus-for-evaluating-language-models-against-professional-codes-of-ethi-2610-29477?lang=esChiosaes2026-10-22T15:13:17ZUna prueba basada en cláusulas revela conflictos con reglas de ética profesionalhttps://chiosa.app/papers/tenetbench-a-clause-anchored-russian-corpus-for-evaluating-language-models-against-professional-codes-of-ethi-2610-29477?lang=frChiosafr2026-10-22T15:13:17ZUn test fondé sur les articles de code révèle des conflits déontologiqueshttps://chiosa.app/papers/the-end-of-consensus-research-2610-30114Chiosaen2026-10-22T15:13:17ZAn AI harness produces a faster consensus protocolhttps://chiosa.app/papers/the-end-of-consensus-research-2610-30114?lang=itChiosait2026-10-22T15:13:17ZUn framework di AI produce un protocollo di consenso più velocehttps://chiosa.app/papers/the-end-of-consensus-research-2610-30114?lang=esChiosaes2026-10-22T15:13:17ZUn entorno de IA produce un protocolo de consenso más rápidohttps://chiosa.app/papers/the-end-of-consensus-research-2610-30114?lang=frChiosafr2026-10-22T15:13:17ZUn cadre d’IA produit un protocole de consensus plus rapidehttps://chiosa.app/papers/uncensored-is-not-unbiased-abliteration-removes-refusals-but-not-script-dependent-bias-about-taiwan-2610-29342Chiosaen2026-10-22T15:13:17ZReducing refusals did not erase Taiwan’s script gaphttps://chiosa.app/papers/uncensored-is-not-unbiased-abliteration-removes-refusals-but-not-script-dependent-bias-about-taiwan-2610-29342?lang=itChiosait2026-10-22T15:13:17ZRidurre i rifiuti non elimina la differenza su Taiwan tra sistemi di scritturahttps://chiosa.app/papers/unlocking-spatial-grounding-in-flow-matching-models-via-snr-tuning-2610-29581Chiosaen2026-10-22T15:13:17ZShifting training helps image models follow coordinateshttps://chiosa.app/papers/unlocking-spatial-grounding-in-flow-matching-models-via-snr-tuning-2610-29581?lang=itChiosait2026-10-22T15:13:17ZSpostare l’addestramento aiuta i modelli a seguire le coordinatehttps://chiosa.app/papers/unlocking-spatial-grounding-in-flow-matching-models-via-snr-tuning-2610-29581?lang=esChiosaes2026-10-22T15:13:17ZCambiar el entrenamiento ayuda a los modelos de imagen a seguir coordenadashttps://chiosa.app/papers/unlocking-spatial-grounding-in-flow-matching-models-via-snr-tuning-2610-29581?lang=frChiosafr2026-10-22T15:13:17ZUn entraînement décalé aide les modèles d’image à suivre les coordonnéeshttps://chiosa.app/papers/what-the-author-wrote-what-the-agent-reads-silent-cue-loss-in-enterprise-document-ingestion-2610-30026Chiosaen2026-10-22T15:13:17ZA document parser preserves visual cues agents need to find requirementshttps://chiosa.app/papers/what-the-author-wrote-what-the-agent-reads-silent-cue-loss-in-enterprise-document-ingestion-2610-30026?lang=itChiosait2026-10-22T15:13:17ZUn parser conserva i segnali visivi utili agli agenti per trovare i requisitihttps://chiosa.app/papers/what-the-author-wrote-what-the-agent-reads-silent-cue-loss-in-enterprise-document-ingestion-2610-30026?lang=esChiosaes2026-10-22T15:13:17ZUn analizador conserva las señales visuales que ayudan a los agentes a encontrar requisitoshttps://chiosa.app/papers/what-the-author-wrote-what-the-agent-reads-silent-cue-loss-in-enterprise-document-ingestion-2610-30026?lang=frChiosafr2026-10-22T15:13:17ZUn analyseur préserve les indices visuels qui aident les agents à repérer les exigenceshttps://chiosa.app/papers/when-thoughts-become-memory-unifying-in-context-learning-and-fine-tuning-2610-29419Chiosaen2026-10-22T15:13:17ZA bit-based method compares context with model weightshttps://chiosa.app/papers/when-thoughts-become-memory-unifying-in-context-learning-and-fine-tuning-2610-29419?lang=itChiosait2026-10-22T15:13:17ZUn metodo basato sui bit confronta il contesto con i pesi del modellohttps://chiosa.app/papers/when-thoughts-become-memory-unifying-in-context-learning-and-fine-tuning-2610-29419?lang=esChiosaes2026-10-22T15:13:17ZUn método basado en bits compara el contexto con los pesos del modelohttps://chiosa.app/papers/when-thoughts-become-memory-unifying-in-context-learning-and-fine-tuning-2610-29419?lang=frChiosafr2026-10-22T15:13:17ZUne méthode en bits compare le contexte aux poids du modèlehttps://chiosa.app/papers/where-should-a-stuck-agent-spend-its-next-tokens-causal-attribution-of-inference-time-resources-2610-28834Chiosaen2026-10-22T15:13:17ZA fixed-state test shows what helps coding agents repair code