https://chiosa.app/papers/anvil-scaling-embodied-spatial-intelligence-through-knowledge-recirculation-2610-21882Chiosaen2026-10-16T11:56:00ZAnvil links spatial learning across related taskshttps://chiosa.app/papers/bitsparse-compact-activation-storage-for-relu-feed-forward-networks-2610-22256Chiosaen2026-10-16T11:56:00ZBitSparse stores sparse feed-forward activations with less memoryhttps://chiosa.app/papers/bitsparse-compact-activation-storage-for-relu-feed-forward-networks-2610-22256?lang=itChiosait2026-10-16T11:56:00ZBitSparse riduce la memoria per archiviare attivazioni feed-forward sparsehttps://chiosa.app/papers/bitsparse-compact-activation-storage-for-relu-feed-forward-networks-2610-22256?lang=esChiosaes2026-10-16T11:56:00ZBitSparse reduce la memoria al guardar activaciones feed-forward dispersashttps://chiosa.app/papers/bitsparse-compact-activation-storage-for-relu-feed-forward-networks-2610-22256?lang=frChiosafr2026-10-16T11:56:00ZBitSparse stocke les activations feed-forward parcimonieuses en utilisant moins de mémoirehttps://chiosa.app/papers/combining-llm-and-dynamic-analysis-for-universal-testing-of-static-application-security-testing-tools-2610-21309Chiosaen2026-10-16T11:56:00ZLLM4SAST finds faults in six Java and Rust code analyzershttps://chiosa.app/papers/flashattention-with-statistics-low-overhead-attention-features-for-compact-hallucination-detectors-2610-21483Chiosaen2026-10-16T11:56:00ZAttention statistics can be extracted without storing full matriceshttps://chiosa.app/papers/intern-decision-efficient-multimodal-decision-making-with-autoregressive-masked-language-modeling-2610-22131Chiosaen2026-10-16T11:56:00ZIntern-Decision scores structured choices in one passhttps://chiosa.app/papers/intern-decision-efficient-multimodal-decision-making-with-autoregressive-masked-language-modeling-2610-22131?lang=itChiosait2026-10-16T11:56:00ZIntern-Decision valuta scelte strutturate in un passaggiohttps://chiosa.app/papers/intern-decision-efficient-multimodal-decision-making-with-autoregressive-masked-language-modeling-2610-22131?lang=esChiosaes2026-10-16T11:56:00ZIntern-Decision puntúa opciones estructuradas en una pasadahttps://chiosa.app/papers/intern-decision-efficient-multimodal-decision-making-with-autoregressive-masked-language-modeling-2610-22131?lang=frChiosafr2026-10-16T11:56:00ZIntern-Decision évalue les choix structurés en un passagehttps://chiosa.app/papers/language-is-a-security-surface-for-agent-guards-2610-22244Chiosaen2026-10-16T11:56:00ZTranslated requests can change agent guards’ safety verdictshttps://chiosa.app/papers/language-is-a-security-surface-for-agent-guards-2610-22244?lang=itChiosait2026-10-16T11:56:00ZLe richieste tradotte possono cambiare i verdetti di sicurezza dei controlli per agentihttps://chiosa.app/papers/language-is-a-security-surface-for-agent-guards-2610-22244?lang=esChiosaes2026-10-16T11:56:00ZLas peticiones traducidas pueden cambiar los veredictos de seguridad de los protectores de agenteshttps://chiosa.app/papers/language-is-a-security-surface-for-agent-guards-2610-22244?lang=frChiosafr2026-10-16T11:56:00ZLes demandes traduites peuvent modifier les verdicts de sécurité des garde-fous d’agentshttps://chiosa.app/papers/multi-sft-permissively-licensed-multilingual-instruction-following-2610-22086Chiosaen2026-10-16T11:56:00ZMulti-SFT builds permissive instruction data across five languageshttps://chiosa.app/papers/opengcua-toward-a-fully-open-stack-for-generalist-computer-use-2610-22200Chiosaen2026-10-16T11:56:00ZOpenGCUA lets computer agents coordinate screen and shell actions in codehttps://chiosa.app/papers/opengcua-toward-a-fully-open-stack-for-generalist-computer-use-2610-22200?lang=esChiosaes2026-10-16T11:56:00ZOpenGCUA permite que agentes informáticos coordinen acciones de pantalla y shell mediante códigohttps://chiosa.app/papers/shortcut-sensitivity-in-agent-safety-evaluations-2610-22243Chiosaen2026-10-16T11:56:00ZA three-level audit tests when safety cues sway guard decisionshttps://chiosa.app/papers/shortcut-sensitivity-in-agent-safety-evaluations-2610-22243?lang=itChiosait2026-10-16T11:56:00ZUn audit in tre livelli verifica quando gli indizi di sicurezza influenzano i guardhttps://chiosa.app/papers/shortcut-sensitivity-in-agent-safety-evaluations-2610-22243?lang=esChiosaes2026-10-16T11:56:00ZUna auditoría en tres niveles prueba cuándo las pistas de seguridad alteran decisiones de guardshttps://chiosa.app/papers/shortcut-sensitivity-in-agent-safety-evaluations-2610-22243?lang=frChiosafr2026-10-16T11:56:00ZUn audit en trois niveaux teste l’influence des indices de sécurité sur les garde-foushttps://chiosa.app/papers/skip-the-compressor-coding-agents-fix-more-issues-when-they-pick-whole-files-2610-21073Chiosaen2026-10-16T11:56:00ZWhole-file context helps coding agents resolve more issueshttps://chiosa.app/papers/spark-0-scaling-spatially-aligned-tactile-pretraining-for-robot-dexterous-manipulation-2610-22342Chiosaen2026-10-16T11:56:00ZRobot touch training improves control with and without touchhttps://chiosa.app/papers/spark-0-scaling-spatially-aligned-tactile-pretraining-for-robot-dexterous-manipulation-2610-22342?lang=itChiosait2026-10-16T11:56:00ZL’addestramento tattile migliora il controllo robotico con e senza tattohttps://chiosa.app/papers/spark-0-scaling-spatially-aligned-tactile-pretraining-for-robot-dexterous-manipulation-2610-22342?lang=esChiosaes2026-10-16T11:56:00ZEl entrenamiento robótico con tacto mejora el control con y sin tactohttps://chiosa.app/papers/spark-0-scaling-spatially-aligned-tactile-pretraining-for-robot-dexterous-manipulation-2610-22342?lang=frChiosafr2026-10-16T11:56:00ZLe toucher aide les robots à manipuler avec ou sans données tactileshttps://chiosa.app/papers/speculative-forcing-for-efficient-autoregressive-video-generation-2610-22347Chiosaen2026-10-16T11:56:00ZVideo generators can stop refining tokens that have settledhttps://chiosa.app/papers/speculative-forcing-for-efficient-autoregressive-video-generation-2610-22347?lang=itChiosait2026-10-16T11:56:00ZI generatori video possono smettere di perfezionare i token stabilihttps://chiosa.app/papers/speculative-forcing-for-efficient-autoregressive-video-generation-2610-22347?lang=esChiosaes2026-10-16T11:56:00ZLos generadores de video pueden dejar de refinar tokens estabilizadoshttps://chiosa.app/papers/surgstreamer-proactive-request-driven-monitoring-of-surgical-video-streams-with-structured-state-and-a-reusab-2610-21895Chiosaen2026-10-16T11:56:00ZSurgStreamer times alerts for events in surgical videohttps://chiosa.app/papers/survicl-a-foundation-model-for-in-context-survival-analysis-2610-22242Chiosaen2026-10-16T11:56:00ZA survival model predicts event curves from censored datahttps://chiosa.app/papers/survicl-a-foundation-model-for-in-context-survival-analysis-2610-22242?lang=itChiosait2026-10-16T11:56:00ZUn modello prevede curve degli eventi da dati censuratihttps://chiosa.app/papers/survicl-a-foundation-model-for-in-context-survival-analysis-2610-22242?lang=esChiosaes2026-10-16T11:56:00ZUn modelo predice curvas de supervivencia con datos censuradoshttps://chiosa.app/papers/when-does-post-call-begin-an-availability-contract-for-earnings-call-nlp-benchmarks-2610-21659Chiosaen2026-10-16T11:56:00ZA reporting framework shows how return timing shifts financial NLP labelshttps://chiosa.app/papers/a-self-improving-agent-that-grows-its-own-dsl-oracle-free-operator-induction-for-exact-question-answering-2610-20809Chiosaen2026-10-16T08:04:15ZA language model grows exact answers by checking new codehttps://chiosa.app/papers/approve-what-exactly-effect-binding-and-trusted-rendering-for-agent-step-up-authorization-2610-21063Chiosaen2026-10-16T08:04:15ZAgent approval can fail when the displayed action is not boundhttps://chiosa.app/papers/comfortably-stuck-evaluation-pitfalls-in-bench2drive-s-closed-loop-protocol-2610-21741Chiosaen2026-10-16T08:04:15ZA driving test’s comfort score can reward staying stillhttps://chiosa.app/papers/comfortably-stuck-evaluation-pitfalls-in-bench2drive-s-closed-loop-protocol-2610-21741?lang=itChiosait2026-10-16T08:04:15ZIl punteggio di comfort di un test di guida può premiare chi resta fermohttps://chiosa.app/papers/compile-the-table-query-calibrated-operator-compression-for-tabular-in-context-learning-2610-11784Chiosaen2026-10-16T08:04:15ZA reusable cache keeps tabular predictions accurate after compressionhttps://chiosa.app/papers/contextlift-breaking-the-spurious-correlation-behind-conditional-emergent-misalignment-2610-21633Chiosaen2026-10-16T08:04:15ZAdding context to safety examples reduces cued misalignmenthttps://chiosa.app/papers/contextlift-breaking-the-spurious-correlation-behind-conditional-emergent-misalignment-2610-21633?lang=itChiosait2026-10-16T08:04:15ZAggiungere contesto agli esempi di sicurezza riduce il disallineamento in risposta agli indizihttps://chiosa.app/papers/contextlift-breaking-the-spurious-correlation-behind-conditional-emergent-misalignment-2610-21633?lang=esChiosaes2026-10-16T08:04:15ZAñadir contexto a ejemplos de seguridad reduce la desalineación ante señaleshttps://chiosa.app/papers/disarm-ing-the-hidden-puppetmaster-an-interface-for-exposing-manipulative-tactics-in-large-language-models-2610-20878Chiosaen2026-10-16T08:04:15ZChat disclosures helped limit belief shifts toward an AI agent’s goalhttps://chiosa.app/papers/dropkv-decoupling-residual-output-perturbation-for-near-optimal-kv-cache-eviction-2610-22130Chiosaen2026-10-16T08:04:15ZDropKV leads tested eviction methods with a small token cachehttps://chiosa.app/papers/dropkv-decoupling-residual-output-perturbation-for-near-optimal-kv-cache-eviction-2610-22130?lang=esChiosaes2026-10-16T08:04:15ZDropKV supera a los métodos evaluados con una caché pequeñahttps://chiosa.app/papers/dropkv-decoupling-residual-output-perturbation-for-near-optimal-kv-cache-eviction-2610-22130?lang=frChiosafr2026-10-16T08:04:15ZDropKV devance les méthodes d’éviction testées avec un cache réduithttps://chiosa.app/papers/emo-a-closed-loop-test-time-computation-framework-via-alternating-context-optimization-and-logits-steering-2610-22160Chiosaen2026-10-16T08:04:15ZA method steers answers as it revises question contexthttps://chiosa.app/papers/emo-a-closed-loop-test-time-computation-framework-via-alternating-context-optimization-and-logits-steering-2610-22160?lang=itChiosait2026-10-16T08:04:15ZUn metodo orienta le risposte mentre rivede il contesto della domandahttps://chiosa.app/papers/emo-a-closed-loop-test-time-computation-framework-via-alternating-context-optimization-and-logits-steering-2610-22160?lang=esChiosaes2026-10-16T08:04:15ZUn método guía las respuestas mientras revisa el contexto de cada preguntahttps://chiosa.app/papers/falcon-fpga-accelerated-latency-certifiable-occupancy-networks-for-camera-based-autonomous-driving-2610-22143Chiosaen2026-10-16T08:04:15ZFixed-work FPGA design bounds two occupancy steps, not the full pipelinehttps://chiosa.app/papers/falcon-fpga-accelerated-latency-certifiable-occupancy-networks-for-camera-based-autonomous-driving-2610-22143?lang=itChiosait2026-10-16T08:04:15ZFPGA a lavoro fisso: tempi definiti per due operazioni, non per l’intera pipelinehttps://chiosa.app/papers/falcon-fpga-accelerated-latency-certifiable-occupancy-networks-for-camera-based-autonomous-driving-2610-22143?lang=esChiosaes2026-10-16T08:04:15ZLa FPGA acota dos pasos de ocupación, no toda la canalizaciónhttps://chiosa.app/papers/falcon-fpga-accelerated-latency-certifiable-occupancy-networks-for-camera-based-autonomous-driving-2610-22143?lang=frChiosafr2026-10-16T08:04:15ZFPGA à charge fixe : deux étapes d’occupation bornées, pas toute la chaînehttps://chiosa.app/papers/fractional-power-of-two-quantization-for-efficient-and-effective-multiplier-free-llm-inference-2610-21257Chiosaen2026-10-16T08:04:15ZA finer weight grid brings multiplier-free LLM inference closerhttps://chiosa.app/papers/imperfect-is-perfect-learned-column-prefilling-for-branch-price-and-cut-2610-21001Chiosaen2026-10-16T08:04:15ZA learned route archive speeds exact vehicle routinghttps://chiosa.app/papers/infuse-a-unified-surface-benchmark-for-indirect-prompt-injection-in-agentic-systems-2610-20776Chiosaen2026-10-16T08:04:15ZINFUSE tests how hidden instructions redirect agentshttps://chiosa.app/papers/lago-language-agents-with-reusable-growing-options-2610-19843Chiosaen2026-10-16T08:04:15ZLAGO reuses specialist agents to solve coding taskshttps://chiosa.app/papers/loud-and-clear-making-visual-evidence-count-in-multimodal-document-rag-2610-21318Chiosaen2026-10-16T08:04:15ZLoud and Clear helps generators use retrieved page evidencehttps://chiosa.app/papers/my-credentials-please-detecting-filesystem-capability-laundering-in-mcp-servers-at-scale-2610-22301Chiosaen2026-10-16T08:04:15ZA framework tests local tools for filesystem leakshttps://chiosa.app/papers/my-credentials-please-detecting-filesystem-capability-laundering-in-mcp-servers-at-scale-2610-22301?lang=itChiosait2026-10-16T08:04:15ZUn framework verifica le fughe di file negli strumenti localihttps://chiosa.app/papers/my-credentials-please-detecting-filesystem-capability-laundering-in-mcp-servers-at-scale-2610-22301?lang=esChiosaes2026-10-16T08:04:15ZUn marco prueba fugas de archivos en herramientas localeshttps://chiosa.app/papers/my-credentials-please-detecting-filesystem-capability-laundering-in-mcp-servers-at-scale-2610-22301?lang=frChiosafr2026-10-16T08:04:15ZUn cadre teste les fuites de fichiers des outils locauxhttps://chiosa.app/papers/noise-space-reward-alignment-for-few-step-generative-models-2610-21125Chiosaen2026-10-16T08:04:15ZA learned noise policy steers fast generators without changing themhttps://chiosa.app/papers/os-helix-unifying-gui-interaction-and-code-execution-in-native-computer-using-agents-2610-22234Chiosaen2026-10-16T08:04:15ZTraining helps desktop agents coordinate clicks and codehttps://chiosa.app/papers/os-helix-unifying-gui-interaction-and-code-execution-in-native-computer-using-agents-2610-22234?lang=esChiosaes2026-10-16T08:04:15ZEl entrenamiento ayuda a coordinar clics y códigohttps://chiosa.app/papers/os-helix-unifying-gui-interaction-and-code-execution-in-native-computer-using-agents-2610-22234?lang=frChiosafr2026-10-16T08:04:15ZL’entraînement aide les agents à combiner interface et codehttps://chiosa.app/papers/outperforming-1000-euclidean-layers-with-hyperbolic-goal-conditioned-rl-2610-20009Chiosaen2026-10-16T08:04:15ZHyperbolic distance helps agents navigate maze routeshttps://chiosa.app/papers/priming-base-models-for-agentic-post-training-an-open-midtraining-recipe-2610-22019Chiosaen2026-10-16T08:04:15ZLong-context midtraining raises selected coding scoreshttps://chiosa.app/papers/priming-base-models-for-agentic-post-training-an-open-midtraining-recipe-2610-22019?lang=itChiosait2026-10-16T08:04:15ZIl midtraining su contesti lunghi alza alcuni punteggi di codinghttps://chiosa.app/papers/priming-base-models-for-agentic-post-training-an-open-midtraining-recipe-2610-22019?lang=esChiosaes2026-10-16T08:04:15ZEl entrenamiento con contexto largo mejora algunas puntuaciones de programaciónhttps://chiosa.app/papers/priming-base-models-for-agentic-post-training-an-open-midtraining-recipe-2610-22019?lang=frChiosafr2026-10-16T08:04:15ZLe midtraining à long contexte améliore certains scores de programmationhttps://chiosa.app/papers/prism-1-position-and-rival-independent-scoring-for-system-one-models-2610-22066Chiosaen2026-10-16T08:04:15ZScoring options separately keeps choices order-stablehttps://chiosa.app/papers/prism-1-position-and-rival-independent-scoring-for-system-one-models-2610-22066?lang=itChiosait2026-10-16T08:04:15ZValutare le opzioni separatamente mantiene stabili le scelte al riordinohttps://chiosa.app/papers/prism-1-position-and-rival-independent-scoring-for-system-one-models-2610-22066?lang=esChiosaes2026-10-16T08:04:15ZPuntuar las opciones por separado mantiene estables las elecciones ante cambios de ordenhttps://chiosa.app/papers/prism-1-position-and-rival-independent-scoring-for-system-one-models-2610-22066?lang=frChiosafr2026-10-16T08:04:15ZÉvaluer les options séparément stabilise les choix face à l’ordrehttps://chiosa.app/papers/rethinking-contrastive-loss-in-clip-post-training-a-complementary-framework-with-frozen-text-encoder-2610-11374Chiosaen2026-10-16T08:04:15ZA smaller contrastive temperature helps CLIP retain accuracyhttps://chiosa.app/papers/risk-controlled-dft-selection-for-machine-learned-interatomic-potentials-2610-19630Chiosaen2026-10-16T08:04:15ZA structure score reduces DFT checks under an error budgethttps://chiosa.app/papers/roleq-role-aligned-scales-for-converting-language-models-to-ternary-2610-21808Chiosaen2026-10-16T08:04:15ZRoleQ aligns ternary weight scales with how model units workhttps://chiosa.app/papers/roleq-role-aligned-scales-for-converting-language-models-to-ternary-2610-21808?lang=itChiosait2026-10-16T08:04:15ZRoleQ adatta le scale dei pesi ternari al ruolo delle unitàhttps://chiosa.app/papers/roleq-role-aligned-scales-for-converting-language-models-to-ternary-2610-21808?lang=esChiosaes2026-10-16T08:04:15ZRoleQ ajusta las escalas de los pesos ternarios a la función de cada unidadhttps://chiosa.app/papers/roleq-role-aligned-scales-for-converting-language-models-to-ternary-2610-21808?lang=frChiosafr2026-10-16T08:04:15ZRoleQ adapte les échelles des poids ternaires au rôle des unitéshttps://chiosa.app/papers/selection-safe-conformal-early-exit-distribution-free-coverage-under-adaptive-depth-2610-20773Chiosaen2026-10-16T08:04:15ZA single calibration protects early-exit prediction setshttps://chiosa.app/papers/shiftqat-quantization-aware-training-for-the-numerical-contract-of-a-fixed-function-edge-cnn-accelerator-2610-21695Chiosaen2026-10-16T08:04:15ZA training flow makes CNNs fit a fixed edge chip’s number ruleshttps://chiosa.app/papers/stateful-testing-for-deep-learning-compilers-2610-22003Chiosaen2026-10-16T08:04:15ZRondo tests compiler bugs that depend on earlier callshttps://chiosa.app/papers/stateful-testing-for-deep-learning-compilers-2610-22003?lang=esChiosaes2026-10-16T08:04:15ZRondo prueba errores del compilador que dependen de llamadas anterioreshttps://chiosa.app/papers/stronger-judges-do-not-see-more-deployment-integrity-blindness-in-llm-evaluation-2610-21135Chiosaen2026-10-16T08:04:15ZOutput scores can miss changes to an AI agent’s runtime stackhttps://chiosa.app/papers/trajectory-level-speculation-with-semantic-verification-for-tool-using-agents-2610-21463Chiosaen2026-10-16T08:04:15ZStrong models can review agent steps instead of writing eachhttps://chiosa.app/papers/transx-technical-report-scaling-sequence-transduction-for-recommendation-2610-20913Chiosaen2026-10-16T08:04:15ZTransX separates recommendation scaling from servinghttps://chiosa.app/papers/what-should-we-remove-evaluating-the-reliability-and-costs-of-rlvr-data-decontamination-2610-21229Chiosaen2026-10-16T08:04:15ZMath overlap rules can remove many distinct problemshttps://chiosa.app/papers/when-the-kv-cache-is-slower-than-the-gpu-scaling-remote-kv-tiers-for-context-parallel-prefill-2610-21500Chiosaen2026-10-16T08:04:15ZWhy a shared prompt cache raised tail latencyhttps://chiosa.app/papers/who-s-behind-the-text-robust-model-authorship-attribution-2610-19583Chiosaen2026-10-16T08:04:15ZATLAS improves model attribution when domains and versions shifthttps://chiosa.app/papers/zoomcap-evidence-routed-visual-zooming-for-recoverable-visual-information-transfer-2610-21320Chiosaen2026-10-16T08:04:15ZZoomCap routes image evidence into richer captionshttps://chiosa.app/papers/anuvani-lid-robust-and-calibrated-spoken-language-identification-for-indian-languages-2610-20263Chiosaen2026-10-15T18:13:38ZDedicated model leads on Indian language speech testshttps://chiosa.app/papers/benchchecker-assessing-the-credibility-of-bug-fixing-benchmarks-for-llms-2610-19447Chiosaen2026-10-15T18:13:38ZA code-history check flags overlap in bug-fixing testshttps://chiosa.app/papers/beyond-maliciousness-static-and-runtime-attack-surface-profiling-of-agent-skills-2610-19264Chiosaen2026-10-15T18:13:38ZProfiling reveals capabilities that threat labels miss in agent skillshttps://chiosa.app/papers/beyond-the-trial-and-error-loop-hybrid-projection-and-automated-tuning-for-distributed-training-2610-19423Chiosaen2026-10-15T18:13:38ZSmall GPU runs can guide large training searcheshttps://chiosa.app/papers/cafe-confining-feature-distillation-to-language-routed-adapters-for-compact-multilingual-automatic-speech-rec-2610-19134Chiosaen2026-10-15T18:13:38ZA speech model confines feature matching to protect Englishhttps://chiosa.app/papers/casi-credit-assignment-via-set-intervention-for-agentic-retrieval-2610-20293Chiosaen2026-10-15T18:13:38ZA method compares search actions to divide shared credithttps://chiosa.app/papers/casi-credit-assignment-via-set-intervention-for-agentic-retrieval-2610-20293?lang=esChiosaes2026-10-15T18:13:38ZUn método compara acciones de búsqueda para repartir el crédito compartidohttps://chiosa.app/papers/counterfactual-gate-support-decomposition-separating-selection-from-redistribution-in-training-data-filtering-2610-19343Chiosaen2026-10-15T18:13:38ZA method separates data selection from support shiftshttps://chiosa.app/papers/delta-tokenizer-reference-identical-tokenization-of-session-deltas-and-segment-parallelism-for-long-prompts-o-2610-19801Chiosaen2026-10-15T18:13:38ZA tokenizer reuses chat history to cut first-token delayshttps://chiosa.app/papers/delta-tokenizer-reference-identical-tokenization-of-session-deltas-and-segment-parallelism-for-long-prompts-o-2610-19801?lang=itChiosait2026-10-15T18:13:38ZUn tokenizer riutilizza la cronologia della chat per ridurre il ritardo al primo tokenhttps://chiosa.app/papers/diagnosing-and-correcting-reward-hacking-in-verifiable-instruction-following-2610-19525Chiosaen2026-10-15T18:13:38ZA calibrated reward helps models follow rules and answer wellhttps://chiosa.app/papers/did-the-llm-answer-the-question-a-semantic-reconstruction-game-for-jailbreak-evaluation-2610-19306Chiosaen2026-10-15T18:13:38ZReconstructing replies steadies jailbreak success reportshttps://chiosa.app/papers/function-space-temporal-self-distillation-for-policy-optimization-2610-19687Chiosaen2026-10-15T18:13:38ZLanguage models preserve useful past predictions with FSPOhttps://chiosa.app/papers/groundgui-gui-grounding-by-high-quality-data-and-distillation-to-reinforcement-learning-2610-19100Chiosaen2026-10-15T18:13:38ZA two-stage training method improves GUI groundinghttps://chiosa.app/papers/how-the-dirty-set-moves-characterizing-schedule-reuse-in-incremental-sparse-matrix-multiplication-2610-19489Chiosaen2026-10-15T18:13:38ZDirect updates usually beat tiled reuse for changing sparse inputshttps://chiosa.app/papers/hybrid-reasoning-agents-2610-18861Chiosaen2026-10-15T18:13:38ZHybridAgent combines latent reasoning with tool use in texthttps://chiosa.app/papers/jetspec-foresight-parallel-drafting-with-self-refinements-2610-19792Chiosaen2026-10-15T18:13:38ZJetSpec++ drafts ahead while the target checks current tokenshttps://chiosa.app/papers/jetspec-foresight-parallel-drafting-with-self-refinements-2610-19792?lang=itChiosait2026-10-15T18:13:38ZJetSpec++ genera la bozza mentre il target controlla i tokenhttps://chiosa.app/papers/jetstream-generating-verified-query-accelerators-for-existing-database-systems-2610-20527Chiosaen2026-10-15T18:13:38ZJetStream speeds up known queries inside existing databaseshttps://chiosa.app/papers/jetstream-generating-verified-query-accelerators-for-existing-database-systems-2610-20527?lang=itChiosait2026-10-15T18:13:38ZJetStream accelera le query note nei database esistentihttps://chiosa.app/papers/jetstream-generating-verified-query-accelerators-for-existing-database-systems-2610-20527?lang=esChiosaes2026-10-15T18:13:38ZJetStream acelera consultas conocidas dentro de bases de datos existenteshttps://chiosa.app/papers/jetstream-generating-verified-query-accelerators-for-existing-database-systems-2610-20527?lang=frChiosafr2026-10-15T18:13:38ZJetStream accélère les requêtes connues dans les bases existanteshttps://chiosa.app/papers/joint-subspace-sparse-attention-for-efficient-long-context-decoding-2610-19026Chiosaen2026-10-15T18:13:38ZSparse attention stays close to full attention at long contexthttps://chiosa.app/papers/manac-1b-instruct-2-reproducible-safety-alignment-for-a-native-brazilian-portuguese-llm-negative-dpo-results-2610-20570Chiosaen2026-10-15T18:13:38ZA Brazilian Portuguese model adds a tested safety stephttps://chiosa.app/papers/manac-1b-instruct-2-reproducible-safety-alignment-for-a-native-brazilian-portuguese-llm-negative-dpo-results-2610-20570?lang=esChiosaes2026-10-15T18:13:38ZUn modelo de portugués brasileño incorpora una etapa de seguridad evaluadahttps://chiosa.app/papers/more-methods-better-agents-a-reality-check-on-self-distillation-in-agentic-reinforcement-learning-2610-19789Chiosaen2026-10-15T18:13:38ZOnly two agent-training methods beat GRPO on averagehttps://chiosa.app/papers/more-methods-better-agents-a-reality-check-on-self-distillation-in-agentic-reinforcement-learning-2610-19789?lang=itChiosait2026-10-15T18:13:38ZSolo due metodi di addestramento degli agenti superano GRPO in mediahttps://chiosa.app/papers/nudge-the-judge-uncovering-program-decision-hijacking-in-real-world-jev-applications-2610-18612Chiosaen2026-10-15T18:13:38ZJevProbe finds frequent decision flips in tested Jev appshttps://chiosa.app/papers/opentwbench-open-taiwan-benchmark-taiwan-domain-knowledge-evaluation-for-large-language-models-and-how-the-me-2610-19848Chiosaen2026-10-15T18:13:38ZOpenTWBench tests Taiwan knowledge and exposes score effectshttps://chiosa.app/papers/opentwbench-open-taiwan-benchmark-taiwan-domain-knowledge-evaluation-for-large-language-models-and-how-the-me-2610-19848?lang=itChiosait2026-10-15T18:13:38ZOpenTWBench valuta le conoscenze su Taiwan e mostra l’effetto dei criteri di punteggiohttps://chiosa.app/papers/p2patch-exploit-guided-vulnerability-repair-2610-20245Chiosaen2026-10-15T18:13:38ZP2Patch finds bypasses and strengthens vulnerability repairshttps://chiosa.app/papers/p2patch-exploit-guided-vulnerability-repair-2610-20245?lang=itChiosait2026-10-15T18:13:38ZP2Patch trova bypass e rafforza le riparazioni delle vulnerabilitàhttps://chiosa.app/papers/p2patch-exploit-guided-vulnerability-repair-2610-20245?lang=esChiosaes2026-10-15T18:13:38ZP2Patch encuentra vías de elusión y refuerza reparaciones de vulnerabilidadeshttps://chiosa.app/papers/p2patch-exploit-guided-vulnerability-repair-2610-20245?lang=frChiosafr2026-10-15T18:13:38ZP2Patch trouve des contournements et renforce les correctifs de sécuritéhttps://chiosa.app/papers/player-1-learning-to-play-from-synthetic-experience-2610-20584Chiosaen2026-10-15T18:13:38ZSynthetic game practice improves Minecraft task successhttps://chiosa.app/papers/player-1-learning-to-play-from-synthetic-experience-2610-20584?lang=itChiosait2026-10-15T18:13:38ZL’esercizio in giochi sintetici migliora il successo in Minecrafthttps://chiosa.app/papers/player-1-learning-to-play-from-synthetic-experience-2610-20584?lang=frChiosafr2026-10-15T18:13:38ZLa pratique dans des jeux synthétiques améliore la réussite aux tâches Minecrafthttps://chiosa.app/papers/prioritizing-microservice-requests-via-retries-2610-20567Chiosaen2026-10-15T18:13:38ZDifferent retry limits help protect important service requestshttps://chiosa.app/papers/rescue-is-this-alarm-worth-acting-on-counterfactual-evaluation-of-vla-runtime-monitors-2610-13516Chiosaen2026-10-15T18:13:38ZRobot alarms can predict failure yet still raise task costshttps://chiosa.app/papers/sheafprune-sheaf-guided-graph-propagation-for-visual-token-pruning-2610-13263Chiosaen2026-10-15T18:13:38ZSheafPrune ranks visual tokens before language decodinghttps://chiosa.app/papers/smart-alone-blind-together-why-accuracy-fails-to-certifytheory-of-mind-in-multi-agent-teams-2610-19051Chiosaen2026-10-15T18:13:38ZWhy shared mistakes can change whether teamwork helpshttps://chiosa.app/papers/tev-scaling-long-horizon-agentic-environments-with-dense-progress-rewards-2610-18868Chiosaen2026-10-15T18:13:38ZLong agent tasks gain verified progress feedbackhttps://chiosa.app/papers/unlocking-structured-pruning-for-reasoning-language-models-via-attention-preserving-mlp-trimming-2610-19121Chiosaen2026-10-15T18:13:38ZLayerTrim compresses reasoning models while preserving their depthhttps://chiosa.app/papers/unsupervised-multimodal-reward-modeling-2610-12707Chiosaen2026-10-15T18:13:38ZA vision-language reward model learns from caption editshttps://chiosa.app/papers/verifying-deployed-bytecode-against-executable-evm-semantics-with-exact-gas-and-no-added-axioms-2610-19764Chiosaen2026-10-15T18:13:38ZBlanc proves properties of deployed Ethereum bytecodehttps://chiosa.app/papers/what-you-sign-is-not-what-you-run-the-verifiability-gap-in-quantized-llm-supply-chains-2610-19498Chiosaen2026-10-15T18:13:38ZSigned model weights do not make quantized files reproduciblehttps://chiosa.app/papers/when-does-momentum-help-in-reinforcement-learning-with-verifiable-rewards-2610-20413Chiosaen2026-10-15T18:13:38ZNear-one momentum improved validation accuracy in two tests