https://chiosa.app/papers/attention-is-only-as-good-as-its-witnesses-when-higher-order-layers-help-and-when-they-do-not-2611-11937Chiosaen2026-11-13T09:42:11ZA path-summing layer transfers graph search to larger graphshttps://chiosa.app/papers/text2ngac-a-multi-llm-pipeline-for-ngac-policy-extraction-with-mutation-based-verification-2611-12572Chiosaen2026-11-13T09:42:11ZA pipeline turns access rules into checked graphshttps://chiosa.app/papers/actor-judge-or-teacher-understanding-decision-models-in-long-horizon-agents-2611-17167Chiosaen2026-11-13T09:39:57ZA weak action chooser can still teach agents which steps helphttps://chiosa.app/papers/auditing-ai-peer-reviewers-a-benchmark-on-real-scientific-papers-2611-17052Chiosaen2026-11-13T09:39:57ZAI reviewers catch paper errors, but results depend on setuphttps://chiosa.app/papers/birddreams-joint-perspective-and-bev-video-generation-for-a-plausible-reactive-driving-world-model-2611-14815Chiosaen2026-11-13T09:39:57ZBirdDreams uses top-down video to model driving motionhttps://chiosa.app/papers/contain-conform-compare-extended-version-a-red-teamed-evaluation-kit-for-small-coding-agents-and-the-full-rec-2611-16485Chiosaen2026-11-13T09:39:57ZA kit makes local coding-agent comparisons clearerhttps://chiosa.app/papers/does-a-self-learning-migration-agent-learn-an-audited-evaluation-of-evidence-graded-experience-memory-for-ora-2611-16143Chiosaen2026-11-13T09:39:57ZA migration agent’s apparent learning fades after leakage checkshttps://chiosa.app/papers/edmd-endpoint-distribution-matching-for-few-step-diffusion-distillation-2611-17194Chiosaen2026-11-13T09:39:57ZMatching final images helps four-step generators improvehttps://chiosa.app/papers/factored-one-sided-adam-moment-for-practical-and-scalable-soap-2611-16905Chiosaen2026-11-13T09:39:57ZA leaner optimizer keeps SOAP’s quality at lower costhttps://chiosa.app/papers/intent-aware-protection-for-ai-agents-2611-16122Chiosaen2026-11-13T09:39:57ZGiving AI agents a purpose profile sharpens security decisionshttps://chiosa.app/papers/isla-a-coordinate-free-measure-weighted-transformer-for-boundary-driven-pdes-2611-17141Chiosaen2026-11-13T09:39:57ZA PDE predictor stays steady when its description changeshttps://chiosa.app/papers/locojax-host-local-coordination-for-small-model-rl-on-tpus-2611-16301Chiosaen2026-11-13T09:39:57ZLoCoJAX cuts step time for small-model TPU traininghttps://chiosa.app/papers/look-here-then-decide-small-decision-models-in-a-harness-find-memory-safety-vulnerabilities-at-scale-on-commo-2611-16565Chiosaen2026-11-13T09:39:57ZA small model helps rank leads to C and C++ memory bugshttps://chiosa.app/papers/megaops-a-kernel-overlapping-framework-with-pytorch-like-modularity-and-megakernel-performance-2611-16248Chiosaen2026-11-13T09:39:57ZMegaOps lets separate GPU kernels overlap through data tileshttps://chiosa.app/papers/on-the-design-of-hyper-connections-simplification-stability-and-scalability-2611-05984Chiosaen2026-11-13T09:39:57ZGrouped residual updates cut training overhead while keeping quality competitivehttps://chiosa.app/papers/physloop-harnessing-llms-physics-awareness-for-self-improving-video-generation-models-2611-16748Chiosaen2026-11-13T09:39:57ZPhysLoop checks simulated physics before refining videoshttps://chiosa.app/papers/resync-residual-latent-state-synchronization-for-deadline-bound-streaming-inference-2611-16523Chiosaen2026-11-13T09:39:57ZA fast updater keeps slow stream models in the loophttps://chiosa.app/papers/retargetable-compilation-for-tensor-kernels-via-a-unified-layout-abstraction-2611-13741Chiosaen2026-11-13T09:39:57ZA tensor-kernel compiler solves layouts across hardwarehttps://chiosa.app/papers/reweave-online-learning-for-workload-adaptive-hnsw-indexes-2611-16060Chiosaen2026-11-13T09:39:57ZA search index rewires links to match query demandhttps://chiosa.app/papers/safety-as-a-budget-calibrating-the-control-channel-of-constrained-grpo-for-vision-language-models-2611-16628Chiosaen2026-11-13T09:39:57ZTraining with direct budgets reduces judged harmful answershttps://chiosa.app/papers/shared-item-level-error-limits-the-aggregation-of-small-language-models-in-fermi-estimation-2611-16315Chiosaen2026-11-13T09:39:57ZShared question errors limit gains from larger model teamshttps://chiosa.app/papers/singularity-bench-can-agents-build-agents-2611-16591Chiosaen2026-11-13T09:39:57ZOpus 5.5 with Claude Code scores above human references on averagehttps://chiosa.app/papers/so-arena-benchmarking-the-capability-spectrum-and-reliability-of-system-one-models-2611-15777Chiosaen2026-11-13T09:39:57ZA new test separates decision skill from reliabilityhttps://chiosa.app/papers/speed-to-power-dominates-energy-costs-for-ai-data-centers-technoeconomic-analysis-of-powering-ai-in-the-conti-2611-17098Chiosaen2026-11-13T09:39:57ZFaster power can matter more than cheap energy for AI centershttps://chiosa.app/papers/stealth-steganographic-text-encoding-agnostic-to-low-level-tensor-hardware-2611-16928Chiosaen2026-11-13T09:39:57ZHidden messages survive mismatched language modelshttps://chiosa.app/papers/step-c-real-time-neural-compression-for-distributed-fiber-optic-sensing-at-scale-2611-17126Chiosaen2026-11-13T09:39:57ZNeural compression keeps live fiber sensing streams flowinghttps://chiosa.app/papers/symmetry-in-time-a-representation-theory-of-linear-recurrent-memory-2611-17162Chiosaen2026-11-13T09:39:57ZFour-dimensional transitions can preserve operation orderhttps://chiosa.app/papers/uav-flow-a-unified-suite-for-benchmarking-and-learning-language-guided-uav-flight-2611-17163Chiosaen2026-11-13T09:39:57ZA drone learns to carry out short flight commandshttps://chiosa.app/papers/unbundling-multimodal-vision-encoding-2611-17334Chiosaen2026-11-13T09:39:57ZMultimodal models need vision capacity, but not a separate encoderhttps://chiosa.app/papers/worlddraft-if-code-builds-the-world-can-video-models-just-render-it-2611-16217Chiosaen2026-11-13T09:39:57ZVideo models trade world preservation for target appearancehttps://chiosa.app/papers/zoomgain-mining-global-to-zoom-perception-gaps-for-fine-grained-visual-supervision-2611-14076Chiosaen2026-11-13T09:39:57ZZoomGain turns zoom-only image details into verified questionshttps://chiosa.app/papers/phrase-level-speculative-decoding-on-a-frozen-target-2611-12464Chiosaen2026-11-12T15:33:05ZPhrase drafting speeds up Qwen3 text generationhttps://chiosa.app/papers/vla-mopd-on-policy-distillation-for-multi-task-vla-policy-consolidation-2611-12286Chiosaen2026-11-12T15:33:05ZA shared robot policy learns from specialist guidancehttps://chiosa.app/papers/when-answer-decodability-is-not-a-decision-separating-final-answer-and-imminent-emission-signals-in-chain-of-2611-11892Chiosaen2026-11-12T15:33:05ZAnswer readouts can track the next candidatehttps://chiosa.app/papers/agent-benchmarks-stop-ranking-before-they-run-out-of-tasks-2611-13541Chiosaen2026-11-12T15:31:16ZAgent benchmarks can stop ranking before tasks run outhttps://chiosa.app/papers/ai-isn-t-getting-dumber-we-re-getting-pickier-rising-standards-in-human-evaluation-of-large-language-models-2611-14333Chiosaen2026-11-12T15:31:16ZVoters increasingly rejected both answers from unchanged AI model pairshttps://chiosa.app/papers/are-dense-rewards-worth-their-tokens-in-grpo-2611-13560Chiosaen2026-11-12T15:31:16ZMatching compute changes which training methods pay offhttps://chiosa.app/papers/bursts-that-outlast-themselves-persistent-degradation-in-prefix-cached-agent-serving-2611-13412Chiosaen2026-11-12T15:31:16ZShort bursts can leave agent-serving caches degradedhttps://chiosa.app/papers/counterfactual-analysis-of-extent-leakage-from-annotation-derived-prompts-in-vision-language-medical-image-se-2611-13970Chiosaen2026-11-12T15:31:16ZMedical captions can give segmentation models lesion sizehttps://chiosa.app/papers/deputybench-evaluating-contextual-safety-in-enterprise-agents-2611-13355Chiosaen2026-11-12T15:31:16ZDeputyBench evaluates enterprise agents’ contextual permission decisionshttps://chiosa.app/papers/dexforesight-grounding-dexterous-policies-with-implicit-spatial-foresight-2611-11166Chiosaen2026-11-12T15:31:16ZDexForesight transfers hand-and-scene foresight into robot actionshttps://chiosa.app/papers/dnavit-discrete-codes-are-general-purpose-visual-representations-2611-12426Chiosaen2026-11-12T15:31:16ZA shared visual code keeps image detail and semantic informationhttps://chiosa.app/papers/don-t-just-trust-your-memory-evidence-linked-memory-reduces-memory-induced-degradation-in-llm-agents-2611-14320Chiosaen2026-11-12T15:31:16ZAgents do better when memory lessons link back to source runshttps://chiosa.app/papers/earlybird-predictive-streaming-speech-recognition-in-29m-parameters-2611-14421Chiosaen2026-11-12T15:31:16ZEarlybird predicts words before speech endshttps://chiosa.app/papers/ensemble-routing-for-prompt-adaptive-cost-aware-llm-inference-2611-10193Chiosaen2026-11-12T15:31:16ZPrompt-aware routing balances model quality and costhttps://chiosa.app/papers/few-step-streaming-video-generation-via-condition-shifting-2611-14033Chiosaen2026-11-12T15:31:16ZAPEX uses one network to generate and estimate streaming videohttps://chiosa.app/papers/forespark-joint-draft-and-verification-budget-scheduling-for-parallel-speculative-decoding-2611-13442Chiosaen2026-11-12T15:31:16ZA serving system cuts draft work that often goes unverifiedhttps://chiosa.app/papers/forward-value-discretization-faster-stochastic-diffusion-sampling-without-high-order-approximation-2611-14328Chiosaen2026-11-12T15:31:16ZA lookahead sampler narrows diffusion’s quality gaphttps://chiosa.app/papers/gtagym-real-aaa-game-tasks-environment-for-evaluation-and-reinforcement-learning-2611-13028Chiosaen2026-11-12T15:31:16ZGTAGym turns GTA V missions into checkable agent taskshttps://chiosa.app/papers/hairuo-forecasting-ocean-eddies-as-fields-and-objects-with-a-satellite-altimetry-foundation-model-2611-14136Chiosaen2026-11-12T15:31:16ZA satellite model forecasts ocean eddies as fields and pathshttps://chiosa.app/papers/incentivizing-self-learning-in-large-language-models-via-reinforcement-learning-2610-39671Chiosaen2026-11-12T15:31:16ZModels can learn from textbooks while proving unfamiliar mathhttps://chiosa.app/papers/macaq-fast-model-agnostic-calibrated-quantization-for-low-precision-weights-and-activations-2611-10649Chiosaen2026-11-12T15:31:16ZMACaQ keeps four-bit quantization within each layerhttps://chiosa.app/papers/mmdm-measuring-massive-multitask-decision-making-2611-14309Chiosaen2026-11-12T15:31:16ZDecision models struggle to recover every valid optionhttps://chiosa.app/papers/one-table-many-queries-benchmarking-generalist-tabular-models-across-generation-imputation-and-prediction-2611-13001Chiosaen2026-11-12T15:31:16ZOne table benchmark reveals shared strengths and trade-offshttps://chiosa.app/papers/pardig-global-precipitation-estimates-learned-from-sparse-satellite-observations-2611-13941Chiosaen2026-11-12T15:31:16ZA global rain map learns from sparse satellite radarhttps://chiosa.app/papers/pulse-or-provenance-what-rppg-based-deepfake-detectors-learn-2611-13992Chiosaen2026-11-12T15:31:16ZDeepfake detectors rely on video cues beyond pulsehttps://chiosa.app/papers/sable-a-small-field-sumcheck-argument-for-bit-exact-inference-of-large-language-models-2611-14083Chiosaen2026-11-12T15:31:16ZSable proves which tokens a quantized language model producedhttps://chiosa.app/papers/scaling-mathematical-proof-supervision-from-structured-textbooks-2610-40394Chiosaen2026-11-12T15:31:16ZA textbook-based corpus links math proofs and chaptershttps://chiosa.app/papers/sinkstate-on-the-fly-svd-of-recurrent-states-for-efficient-hybrid-llm-serving-2611-14246Chiosaen2026-11-12T15:31:16ZA compact state speeds up hybrid model servinghttps://chiosa.app/papers/steerdistill-quantizing-llm-agents-via-closed-loop-teacher-steering-2611-13410Chiosaen2026-11-12T15:31:16ZTeacher-guided corrections help quantized agents recoverhttps://chiosa.app/papers/swanada-split-encoder-capability-routed-conformer-ctc-and-sub-30-mb-on-device-speech-recognition-across-twelv-2611-13814Chiosaen2026-11-12T15:31:16ZSwanada shares speech-recognition models across 12 languageshttps://chiosa.app/papers/tree-ising-repairing-language-models-with-bit-level-ising-edits-2611-14210Chiosaen2026-11-12T15:31:16ZA sparse weight repair improves math answers from a looping model