https://chiosa.app/papers/agreement-without-alignment-label-free-rubric-clarification-increases-llm-judge-consensus-but-not-agreement-w-2610-13303Chiosaen2026-10-13T17:08:19ZJudge disagreement clarifies rubrics but not human alignmenthttps://chiosa.app/papers/agreement-without-alignment-label-free-rubric-clarification-increases-llm-judge-consensus-but-not-agreement-w-2610-13303?lang=itChiosait2026-10-13T17:08:19ZI disaccordi tra valutatori chiariscono le rubriche, ma non l’accordo con gli espertihttps://chiosa.app/papers/agreement-without-alignment-label-free-rubric-clarification-increases-llm-judge-consensus-but-not-agreement-w-2610-13303?lang=esChiosaes2026-10-13T17:08:19ZLos desacuerdos entre jueces aclaran rúbricas, pero no mejoran la coincidencia humanahttps://chiosa.app/papers/agreement-without-alignment-label-free-rubric-clarification-increases-llm-judge-consensus-but-not-agreement-w-2610-13303?lang=frChiosafr2026-10-13T17:08:19ZLes désaccords entre juges clarifient les barèmes, pas leur proximité avec les humainshttps://chiosa.app/papers/apex-harm-evaluating-agent-safety-in-shared-professional-workspaces-2610-13623Chiosaen2026-10-13T17:08:19ZA new test separates seeing an attack from carrying it outhttps://chiosa.app/papers/apex-harm-evaluating-agent-safety-in-shared-professional-workspaces-2610-13623?lang=itChiosait2026-10-13T17:08:19ZUn nuovo test distingue tra vedere un attacco e metterlo in attohttps://chiosa.app/papers/apex-harm-evaluating-agent-safety-in-shared-professional-workspaces-2610-13623?lang=frChiosafr2026-10-13T17:08:19ZUn nouveau test distingue le fait de voir une attaque de celui de la mener à bienhttps://chiosa.app/papers/duorate-two-rate-per-expert-precision-allocation-for-mixture-of-experts-serving-2610-13328Chiosaen2026-10-13T17:08:19ZDuoRate lets MoE models store and read weights at different rateshttps://chiosa.app/papers/duorate-two-rate-per-expert-precision-allocation-for-mixture-of-experts-serving-2610-13328?lang=itChiosait2026-10-13T17:08:19ZDuoRate permette ai modelli MoE di memorizzare e leggere i pesi a misure diversehttps://chiosa.app/papers/duorate-two-rate-per-expert-precision-allocation-for-mixture-of-experts-serving-2610-13328?lang=esChiosaes2026-10-13T17:08:19ZDuoRate permite a los modelos MoE almacenar y leer pesos a tasas distintashttps://chiosa.app/papers/ember-making-unified-virtual-memory-practical-for-discrete-memory-gpus-2610-13408Chiosaen2026-10-13T17:08:19ZEmber keeps oversized GPU workloads moving with adaptive placementhttps://chiosa.app/papers/improving-semantic-coverage-in-ai-test-evaluation-through-generative-augmentation-2610-13697Chiosaen2026-10-13T17:08:19ZSynthetic night images change test-set model rankingshttps://chiosa.app/papers/improving-semantic-coverage-in-ai-test-evaluation-through-generative-augmentation-2610-13697?lang=itChiosait2026-10-13T17:08:19ZLe immagini notturne sintetiche cambiano la classifica dei modelli sul set di testhttps://chiosa.app/papers/inject-a-turn-not-a-token-learning-where-to-redirect-llm-reasoning-for-self-correction-2610-13672Chiosaen2026-10-13T17:08:19ZA new user turn helps redirect failed reasoninghttps://chiosa.app/papers/inject-a-turn-not-a-token-learning-where-to-redirect-llm-reasoning-for-self-correction-2610-13672?lang=itChiosait2026-10-13T17:08:19ZUn nuovo messaggio dell’utente reindirizza il ragionamento erratohttps://chiosa.app/papers/inject-a-turn-not-a-token-learning-where-to-redirect-llm-reasoning-for-self-correction-2610-13672?lang=esChiosaes2026-10-13T17:08:19ZUn nuevo turno del usuario ayuda a redirigir el razonamiento fallidohttps://chiosa.app/papers/inject-a-turn-not-a-token-learning-where-to-redirect-llm-reasoning-for-self-correction-2610-13672?lang=frChiosafr2026-10-13T17:08:19ZUn nouveau message aide à réorienter un raisonnement défaillanthttps://chiosa.app/papers/instantfusion-a-shared-latent-interface-for-heterogeneous-image-generators-2610-13944Chiosaen2026-10-13T17:08:19ZA shared interface lets image generators exchange denoising stateshttps://chiosa.app/papers/lapmerge-differentiable-exact-budget-token-merging-in-linear-time-2610-13858Chiosaen2026-10-13T17:08:19ZLapMerge makes ordered token merging learnablehttps://chiosa.app/papers/lapmerge-differentiable-exact-budget-token-merging-in-linear-time-2610-13858?lang=itChiosait2026-10-13T17:08:19ZLapMerge rende apprendibile il merge di token ordinatihttps://chiosa.app/papers/lapmerge-differentiable-exact-budget-token-merging-in-linear-time-2610-13858?lang=esChiosaes2026-10-13T17:08:19ZLapMerge permite aprender a combinar tokens ordenadoshttps://chiosa.app/papers/maneuverflow-maneuver-hypothesis-generation-with-future-evidence-for-trajectory-evaluation-and-refinement-in-2610-13409Chiosaen2026-10-13T17:08:19ZA driving planner organizes candidate paths by motionhttps://chiosa.app/papers/maneuverflow-maneuver-hypothesis-generation-with-future-evidence-for-trajectory-evaluation-and-refinement-in-2610-13409?lang=itChiosait2026-10-13T17:08:19ZUn pianificatore di guida organizza i percorsi candidati per movimentohttps://chiosa.app/papers/maneuverflow-maneuver-hypothesis-generation-with-future-evidence-for-trajectory-evaluation-and-refinement-in-2610-13409?lang=esChiosaes2026-10-13T17:08:19ZUn planificador de conducción organiza rutas candidatas por movimientohttps://chiosa.app/papers/maneuverflow-maneuver-hypothesis-generation-with-future-evidence-for-trajectory-evaluation-and-refinement-in-2610-13409?lang=frChiosafr2026-10-13T17:08:19ZUn planificateur de conduite organise les trajectoires candidates par mouvementhttps://chiosa.app/papers/on-policy-or-off-policy-learning-a-systematic-study-of-distillation-dynamics-2609-35259Chiosaen2026-10-13T17:08:19ZOn-policy distillation has no consistent edge in core testshttps://chiosa.app/papers/on-policy-or-off-policy-learning-a-systematic-study-of-distillation-dynamics-2609-35259?lang=itChiosait2026-10-13T17:08:19ZLa distillazione on-policy non offre un vantaggio costante nei test chiavehttps://chiosa.app/papers/on-policy-or-off-policy-learning-a-systematic-study-of-distillation-dynamics-2609-35259?lang=frChiosafr2026-10-13T17:08:19ZLa distillation on-policy n’a pas d’avantage systématique aux tests cléshttps://chiosa.app/papers/proactive-agentic-orchestration-2610-13485Chiosaen2026-10-13T17:08:19ZPOrchestra lets agents revise task setups mid-taskhttps://chiosa.app/papers/proactive-agentic-orchestration-2610-13485?lang=frChiosafr2026-10-13T17:08:19ZPOrchestra permet aux agents de modifier leur configuration en cours de tâchehttps://chiosa.app/papers/reins-agent-guided-thread-scheduling-for-concurrency-bug-detection-and-reproduction-2610-13421Chiosaen2026-10-13T17:08:19ZReins steers thread schedules toward concurrency bugshttps://chiosa.app/papers/reins-agent-guided-thread-scheduling-for-concurrency-bug-detection-and-reproduction-2610-13421?lang=esChiosaes2026-10-13T17:08:19ZReins orienta las planificaciones de hilos hacia errores de concurrenciahttps://chiosa.app/papers/replayssm-cache-ssm-inputs-not-state-2610-13778Chiosaen2026-10-13T17:08:19ZReplaySSM cuts recurrent-state traffic during language-model decodinghttps://chiosa.app/papers/replayssm-cache-ssm-inputs-not-state-2610-13778?lang=itChiosait2026-10-13T17:08:19ZReplaySSM riduce il traffico dello stato ricorrente nel decoding dei modelli linguisticihttps://chiosa.app/papers/replayssm-cache-ssm-inputs-not-state-2610-13778?lang=esChiosaes2026-10-13T17:08:19ZReplaySSM reduce el tráfico del estado recurrente al decodificar modelos de lenguajehttps://chiosa.app/papers/replayssm-cache-ssm-inputs-not-state-2610-13778?lang=frChiosafr2026-10-13T17:08:19ZReplaySSM réduit le trafic d’état récurrent au décodage des modèles de langagehttps://chiosa.app/papers/rrsi-regularized-recursive-self-improvement-of-agent-harnesses-2609-24972Chiosaen2026-10-13T17:08:19ZA harness-editing method improves transfer to unseen taskshttps://chiosa.app/papers/rrsi-regularized-recursive-self-improvement-of-agent-harnesses-2609-24972?lang=itChiosait2026-10-13T17:08:19ZUn metodo per modificare gli harness migliora il trasferimento ad attività mai vistehttps://chiosa.app/papers/rrsi-regularized-recursive-self-improvement-of-agent-harnesses-2609-24972?lang=esChiosaes2026-10-13T17:08:19ZUn método para modificar arneses mejora la transferencia a tareas nuevashttps://chiosa.app/papers/shortcut-learning-in-geospatial-foundation-model-embeddings-for-cross-disaster-damage-assessment-2610-13766Chiosaen2026-10-13T17:08:19ZDisaster damage scores drop when a whole event is held outhttps://chiosa.app/papers/shortcut-learning-in-geospatial-foundation-model-embeddings-for-cross-disaster-damage-assessment-2610-13766?lang=esChiosaes2026-10-13T17:08:19ZLas puntuaciones de daños bajan al dejar fuera un desastre completohttps://chiosa.app/papers/taking-tokenization-off-the-host-for-agentic-llm-serving-2610-13325Chiosaen2026-10-13T17:08:19ZA GPU tokenizer preserves IDs while speeding cached requestshttps://chiosa.app/papers/taking-tokenization-off-the-host-for-agentic-llm-serving-2610-13325?lang=itChiosait2026-10-13T17:08:19ZUn tokenizer su GPU accelera le richieste con cachehttps://chiosa.app/papers/taking-tokenization-off-the-host-for-agentic-llm-serving-2610-13325?lang=esChiosaes2026-10-13T17:08:19ZUn tokenizador para GPU conserva los identificadores y acelera solicitudes con cachéhttps://chiosa.app/papers/taking-tokenization-off-the-host-for-agentic-llm-serving-2610-13325?lang=frChiosafr2026-10-13T17:08:19ZUn tokeniseur pour GPU conserve les identifiants et accélère les requêtes en cachehttps://chiosa.app/papers/the-limits-of-query-monitoring-against-distributed-model-extraction-2610-13083Chiosaen2026-10-13T17:08:19ZA query-only monitor finds coordinated accounts despite prompt paddinghttps://chiosa.app/papers/the-limits-of-query-monitoring-against-distributed-model-extraction-2610-13083?lang=itChiosait2026-10-13T17:08:19ZUn monitor basato solo sulle query individua account coordinati nonostante i prompt aggiuntihttps://chiosa.app/papers/when-should-a-software-factory-be-lights-off-benchmarking-code-review-agents-on-bug-detection-2610-13668Chiosaen2026-10-13T17:08:19ZA bug-based test finds limits in automated code reviewhttps://chiosa.app/papers/when-should-a-software-factory-be-lights-off-benchmarking-code-review-agents-on-bug-detection-2610-13668?lang=itChiosait2026-10-13T17:08:19ZUn test basato sui bug rivela i limiti della code review automaticahttps://chiosa.app/papers/when-should-a-software-factory-be-lights-off-benchmarking-code-review-agents-on-bug-detection-2610-13668?lang=esChiosaes2026-10-13T17:08:19ZUna prueba basada en fallos revela límites de la revisión automática de códigohttps://chiosa.app/papers/when-should-a-software-factory-be-lights-off-benchmarking-code-review-agents-on-bug-detection-2610-13668?lang=frChiosafr2026-10-13T17:08:19ZUn test fondé sur les bugs révèle les limites de la revue de code automatiséehttps://chiosa.app/papers/x-avdd-cross-attentive-audio-visual-dataset-distillation-2610-13687Chiosaen2026-10-13T17:08:19ZTwo-way attention helps compact audio-visual data stay usefulhttps://chiosa.app/papers/x-avdd-cross-attentive-audio-visual-dataset-distillation-2610-13687?lang=itChiosait2026-10-13T17:08:19ZLa cross-attention bidirezionale aiuta i dati audio-visivi compatti a restare utilihttps://chiosa.app/papers/x-avdd-cross-attentive-audio-visual-dataset-distillation-2610-13687?lang=esChiosaes2026-10-13T17:08:19ZLa atención bidireccional ayuda a conservar la utilidad de datos audiovisuales compactoshttps://chiosa.app/papers/x-avdd-cross-attentive-audio-visual-dataset-distillation-2610-13687?lang=frChiosafr2026-10-13T17:08:19ZL’attention bidirectionnelle aide les données audio-visuelles compactes à rester utileshttps://chiosa.app/papers/agentvalve-sybil-resilient-payment-authorization-for-permissionless-ai-markets-2610-12948Chiosaen2026-10-13T08:02:27ZAgentValve caps Sybil-driven payment exposure by trust capacityhttps://chiosa.app/papers/agentvalve-sybil-resilient-payment-authorization-for-permissionless-ai-markets-2610-12948?lang=esChiosaes2026-10-13T08:02:27ZAgentValve limita la exposición a pagos Sybil con capacidad de confianzahttps://chiosa.app/papers/agentvalve-sybil-resilient-payment-authorization-for-permissionless-ai-markets-2610-12948?lang=frChiosafr2026-10-13T08:02:27ZAgentValve borne l’exposition aux paiements liée aux identités Sybil grâce à la capacité de confiancehttps://chiosa.app/papers/arc-a-reasoning-recipe-for-robot-foundation-models-2610-12386Chiosaen2026-10-13T08:02:27ZRobot models improve when actions explain intended effectshttps://chiosa.app/papers/arc-a-reasoning-recipe-for-robot-foundation-models-2610-12386?lang=itChiosait2026-10-13T08:02:27ZI modelli robotici migliorano quando le azioni spiegano l’effetto desideratohttps://chiosa.app/papers/arc-a-reasoning-recipe-for-robot-foundation-models-2610-12386?lang=esChiosaes2026-10-13T08:02:27ZLos modelos robóticos mejoran cuando sus acciones explican el efecto deseadohttps://chiosa.app/papers/co-modifying-application-code-and-deployment-infrastructure-a-causal-study-of-ai-agent-pull-requests-2610-12849Chiosaen2026-10-13T08:02:27ZA framework shows where AI infrastructure changes slow mergeshttps://chiosa.app/papers/contextual-correction-of-parallel-draftsfor-speculative-decoding-2610-12570Chiosaen2026-10-13T08:02:27ZParallel draft correction raises decoding speed without sequential repairhttps://chiosa.app/papers/contextual-correction-of-parallel-draftsfor-speculative-decoding-2610-12570?lang=itChiosait2026-10-13T08:02:27ZCorreggere le bozze parallele aumenta la velocità di decodifica senza correzione sequenzialehttps://chiosa.app/papers/designing-protein-structures-from-natural-language-2610-13917Chiosaen2026-10-13T08:02:27ZOne language interface steers protein structure and binder designhttps://chiosa.app/papers/diffusionart-efficient-stable-and-scalable-reinforcement-of-diffusion-models-2610-12572Chiosaen2026-10-13T08:02:27ZA visual-generation trainer makes reward learning faster and steadierhttps://chiosa.app/papers/diffusionart-efficient-stable-and-scalable-reinforcement-of-diffusion-models-2610-12572?lang=esChiosaes2026-10-13T08:02:27ZUn método de entrenamiento visual aprende recompensas más rápido y con más estabilidadhttps://chiosa.app/papers/diffusionart-efficient-stable-and-scalable-reinforcement-of-diffusion-models-2610-12572?lang=frChiosafr2026-10-13T08:02:27ZUne méthode accélère et stabilise l’entraînement des générateurs visuelshttps://chiosa.app/papers/distressbench-evaluating-multi-dimensional-crisis-support-in-large-language-models-2610-12758Chiosaen2026-10-13T08:02:27ZA crisis-support test shows recognition is not enoughhttps://chiosa.app/papers/distressbench-evaluating-multi-dimensional-crisis-support-in-large-language-models-2610-12758?lang=esChiosaes2026-10-13T08:02:27ZUna prueba de apoyo en crisis muestra que reconocer no bastahttps://chiosa.app/papers/distressbench-evaluating-multi-dimensional-crisis-support-in-large-language-models-2610-12758?lang=frChiosafr2026-10-13T08:02:27ZUn test montre que reconnaître une crise ne suffit pashttps://chiosa.app/papers/dream-to-remember-continual-steering-of-generalist-robot-policies-with-world-models-2610-13816Chiosaen2026-10-13T08:02:27ZWorld-model rehearsal helps robots retain skills while learninghttps://chiosa.app/papers/dream-to-remember-continual-steering-of-generalist-robot-policies-with-world-models-2610-13816?lang=esChiosaes2026-10-13T08:02:27ZEl repaso con modelos del mundo ayuda a los robots a conservar habilidadeshttps://chiosa.app/papers/dream-to-remember-continual-steering-of-generalist-robot-policies-with-world-models-2610-13816?lang=frChiosafr2026-10-13T08:02:27ZLa répétition par modèle du monde aide les robots à préserver leurs compétenceshttps://chiosa.app/papers/dualquant-dual-scale-calibration-free-weight-and-activation-quantization-for-llms-2610-13767Chiosaen2026-10-13T08:02:27ZDualQuant quantizes language models without calibration datahttps://chiosa.app/papers/dualquant-dual-scale-calibration-free-weight-and-activation-quantization-for-llms-2610-13767?lang=itChiosait2026-10-13T08:02:27ZDualQuant quantizza i modelli linguistici senza dati di calibrazionehttps://chiosa.app/papers/dualquant-dual-scale-calibration-free-weight-and-activation-quantization-for-llms-2610-13767?lang=esChiosaes2026-10-13T08:02:27ZDualQuant cuantiza modelos de lenguaje sin datos de calibraciónhttps://chiosa.app/papers/dualquant-dual-scale-calibration-free-weight-and-activation-quantization-for-llms-2610-13767?lang=frChiosafr2026-10-13T08:02:27ZDualQuant quantifie les modèles de langage sans données de calibrationhttps://chiosa.app/papers/forte-a-production-framework-for-feature-evaluation-and-deprecation-in-industrial-recommender-systems-2610-13806Chiosaen2026-10-13T08:02:27ZFORTE turns feature removal into a staged production workflowhttps://chiosa.app/papers/images-talk-back-guiding-mmdits-by-suppressing-image-to-text-feedback-2610-13941Chiosaen2026-10-13T08:02:27ZSuppressing image-to-text feedback improves generationhttps://chiosa.app/papers/images-talk-back-guiding-mmdits-by-suppressing-image-to-text-feedback-2610-13941?lang=esChiosaes2026-10-13T08:02:27ZSuprimir la retroalimentación de imagen a texto mejora la generaciónhttps://chiosa.app/papers/lapt-scale-scaling-embodied-policies-with-latent-action-pre-training-on-human-videos-2610-13732Chiosaen2026-10-13T08:02:27ZHuman videos can improve robot policies after alignmenthttps://chiosa.app/papers/lapt-scale-scaling-embodied-policies-with-latent-action-pre-training-on-human-videos-2610-13732?lang=itChiosait2026-10-13T08:02:27ZI video umani possono migliorare le policy robotiche dopo l'allineamentohttps://chiosa.app/papers/lapt-scale-scaling-embodied-policies-with-latent-action-pre-training-on-human-videos-2610-13732?lang=esChiosaes2026-10-13T08:02:27ZLos videos humanos pueden mejorar políticas robóticas tras la alineaciónhttps://chiosa.app/papers/locate-judge-read-role-specialized-tool-agents-over-a-knowledge-point-index-for-open-domain-multimodal-docume-2610-13128Chiosaen2026-10-13T08:02:27ZSeparating search from reading improves document question answeringhttps://chiosa.app/papers/locate-judge-read-role-specialized-tool-agents-over-a-knowledge-point-index-for-open-domain-multimodal-docume-2610-13128?lang=esChiosaes2026-10-13T08:02:27ZSeparar la búsqueda de la lectura mejora las respuestas sobre documentoshttps://chiosa.app/papers/locate-judge-read-role-specialized-tool-agents-over-a-knowledge-point-index-for-open-domain-multimodal-docume-2610-13128?lang=frChiosafr2026-10-13T08:02:27ZSéparer recherche et lecture améliore les réponses sur les documentshttps://chiosa.app/papers/nomos-compiling-written-policies-into-statically-verified-tool-call-gates-for-llm-agents-2610-11030Chiosaen2026-10-13T08:02:27ZNOMOS makes written policies safer for tool-using agentshttps://chiosa.app/papers/nomos-compiling-written-policies-into-statically-verified-tool-call-gates-for-llm-agents-2610-11030?lang=esChiosaes2026-10-13T08:02:27ZNOMOS hace más seguras las políticas escritas para agentes con herramientashttps://chiosa.app/papers/nomos-compiling-written-policies-into-statically-verified-tool-call-gates-for-llm-agents-2610-11030?lang=frChiosafr2026-10-13T08:02:27ZNOMOS rend les politiques écrites plus sûres pour les agents qui utilisent des outilshttps://chiosa.app/papers/o-do-open-vocabulary-speech-recognition-on-a-5-microcontroller-without-a-neural-accelerator-2610-13447Chiosaen2026-10-13T08:02:27ZOpen-vocabulary speech runs on a low-cost microcontrollerhttps://chiosa.app/papers/one-model-to-loop-them-all-elastic-latent-recurrence-for-language-models-2610-12582Chiosaen2026-10-13T08:02:27ZOne language model gains accuracy by spending more inference loopshttps://chiosa.app/papers/quilt-rethinking-sparse-attention-prefill-through-shared-query-execution-2610-11134Chiosaen2026-10-13T08:02:27ZQUILT speeds long-context prefill by sharing query workhttps://chiosa.app/papers/quilt-rethinking-sparse-attention-prefill-through-shared-query-execution-2610-11134?lang=itChiosait2026-10-13T08:02:27ZQUILT accelera il prefill con contesto lungo condividendo il lavoro tra queryhttps://chiosa.app/papers/quilt-rethinking-sparse-attention-prefill-through-shared-query-execution-2610-11134?lang=esChiosaes2026-10-13T08:02:27ZQUILT acelera el prellenado de contextos largos al compartir trabajo entre consultashttps://chiosa.app/papers/quilt-rethinking-sparse-attention-prefill-through-shared-query-execution-2610-11134?lang=frChiosafr2026-10-13T08:02:27ZQUILT accélère le prefill des longs contextes en partageant le travail des requêteshttps://chiosa.app/papers/read-what-matters-query-adaptive-quantization-for-kv-caches-2610-11245Chiosaen2026-10-13T08:02:27ZReadKV keeps cache detail without reading it allhttps://chiosa.app/papers/reflex-visually-grounded-reactive-humanoid-control-2610-12772Chiosaen2026-10-13T08:02:27ZReflex helps a humanoid catch thrown boxes with moving-camera visionhttps://chiosa.app/papers/reflex-visually-grounded-reactive-humanoid-control-2610-12772?lang=itChiosait2026-10-13T08:02:27ZReflex aiuta un umanoide ad afferrare scatole con una telecamera mobilehttps://chiosa.app/papers/reflex-visually-grounded-reactive-humanoid-control-2610-12772?lang=frChiosafr2026-10-13T08:02:27ZReflex aide un humanoïde à attraper des boîtes grâce à sa caméra mobilehttps://chiosa.app/papers/skillopt-executive-strategy-for-self-evolving-agent-skills-2605-23904Chiosaen2026-10-13T08:02:27ZWritten skills improve agent procedures without changing weightshttps://chiosa.app/papers/staple-static-sparse-upcycling-of-pretrained-checkpoints-via-embedding-modules-2610-12599Chiosaen2026-10-13T08:02:27ZStatic memory expands pretrained models without sparse pretraininghttps://chiosa.app/papers/staple-static-sparse-upcycling-of-pretrained-checkpoints-via-embedding-modules-2610-12599?lang=itChiosait2026-10-13T08:02:27ZLa memoria statica amplia i modelli preaddestrati senza preaddestramento sparsohttps://chiosa.app/papers/tabular-foundation-models-are-effectively-shallow-2610-13253Chiosaen2026-10-13T08:02:27ZTabular transformers can replace half their blocks with low classification losshttps://chiosa.app/papers/testprism-rethinking-test-evaluation-beyond-a-single-reference-2610-12289Chiosaen2026-10-13T08:02:27ZGenerated tests often fail across valid implementationshttps://chiosa.app/papers/testprism-rethinking-test-evaluation-beyond-a-single-reference-2610-12289?lang=itChiosait2026-10-13T08:02:27ZI test generati spesso falliscono con implementazioni validehttps://chiosa.app/papers/tofu-exploiting-and-fuzzing-the-llm-orchestrator-tool-trust-boundary-2610-13651Chiosaen2026-10-13T08:02:27ZSecurity bugs can hide inside valid LLM tool callshttps://chiosa.app/papers/tofu-exploiting-and-fuzzing-the-llm-orchestrator-tool-trust-boundary-2610-13651?lang=itChiosait2026-10-13T08:02:27ZBug di sicurezza possono nascondersi in chiamate valide agli strumenti degli LLMhttps://chiosa.app/papers/tofu-exploiting-and-fuzzing-the-llm-orchestrator-tool-trust-boundary-2610-13651?lang=esChiosaes2026-10-13T08:02:27ZLos fallos de seguridad pueden ocultarse en llamadas válidas a herramientas de LLMhttps://chiosa.app/papers/training-on-the-future-a-delay-aware-audit-of-test-time-adaptation-for-time-series-forecasting-2610-12232Chiosaen2026-10-13T08:02:27ZWhy delayed labels change forecasting adaptation resultshttps://chiosa.app/papers/training-on-the-future-a-delay-aware-audit-of-test-time-adaptation-for-time-series-forecasting-2610-12232?lang=itChiosait2026-10-13T08:02:27ZPerché le etichette ritardate cambiano i risultati dell’adattamento nelle previsionihttps://chiosa.app/papers/training-on-the-future-a-delay-aware-audit-of-test-time-adaptation-for-time-series-forecasting-2610-12232?lang=frChiosafr2026-10-13T08:02:27ZPourquoi les étiquettes tardives changent les résultats de l’adaptation des prévisionshttps://chiosa.app/papers/understanding-and-mitigating-cross-session-attacks-via-persistent-artifacts-in-agent-skill-composition-2610-13227Chiosaen2026-10-13T08:02:27ZPersistent artifacts let attacks cross agent sessionshttps://chiosa.app/papers/understanding-and-mitigating-cross-session-attacks-via-persistent-artifacts-in-agent-skill-composition-2610-13227?lang=esChiosaes2026-10-13T08:02:27ZLos artefactos persistentes permiten ataques entre sesiones de agenteshttps://chiosa.app/papers/understanding-and-mitigating-cross-session-attacks-via-persistent-artifacts-in-agent-skill-composition-2610-13227?lang=frChiosafr2026-10-13T08:02:27ZLes artefacts persistants font passer les attaques d’une session d’agent à l’autrehttps://chiosa.app/papers/what-counts-as-evidence-epistemic-cue-utilization-in-large-language-models-2610-12788Chiosaen2026-10-13T08:02:27ZSome models spot reliable sources without using that cuehttps://chiosa.app/papers/what-counts-as-evidence-epistemic-cue-utilization-in-large-language-models-2610-12788?lang=itChiosait2026-10-13T08:02:27ZAlcuni modelli riconoscono le fonti affidabili ma non usano quell’indiziohttps://chiosa.app/papers/what-counts-as-evidence-epistemic-cue-utilization-in-large-language-models-2610-12788?lang=esChiosaes2026-10-13T08:02:27ZAlgunos modelos detectan fuentes fiables, pero no usan esa señalhttps://chiosa.app/papers/what-counts-as-evidence-epistemic-cue-utilization-in-large-language-models-2610-12788?lang=frChiosafr2026-10-13T08:02:27ZCertains modèles repèrent les sources fiables sans tenir compte de leur historique