https://aiunderstanding.org/news/google-research-reports-mobility-informed-embeddings-improve-ai-predictions-about https://aiunderstanding.org/api/news/image/google-research-reports-mobility-informed-embeddings-improve-ai-predictions-about An empty Houston commercial street at dawn with shuttered storefronts, a bus shelter, parked cars, and concrete sidewalks. AI Understanding en 2026-08-21T18:48:35.397Z Google Research reports mobility-informed embeddings improve AI predictions about places https://aiunderstanding.org/news/paper-theorizes-how-ai-interactions-could-narrow-people-s-self-understanding https://aiunderstanding.org/api/news/image/paper-theorizes-how-ai-interactions-could-narrow-people-s-self-understanding A featureless white humanoid robot standing in an empty interaction laboratory beside a chair and table in soft daylight. AI Understanding en 2026-08-21T18:38:24.587Z Paper theorizes how AI interactions could narrow people’s self-understanding https://aiunderstanding.org/news/anthropic-expands-claude-mythos-5-for-cyber-defense-and-pledges-35-million-in-open https://aiunderstanding.org/api/news/image/anthropic-expands-claude-mythos-5-for-cyber-defense-and-pledges-35-million-in-open An unoccupied server room at a small open-source software foundation, with secured equipment and network cables arranged on a metal workbench. AI Understanding en 2026-08-21T17:56:39.433Z Anthropic expands Claude Mythos 5 for cyber defense and pledges $35 million in open-source credits https://aiunderstanding.org/news/google-research-describes-an-ai-system-for-prioritizing-biomarkers-from-wearable-d https://aiunderstanding.org/api/news/image/google-research-describes-an-ai-system-for-prioritizing-biomarkers-from-wearable-d Researchers in a medical laboratory with wearable sensor equipment arranged on a workbench. AI Understanding en 2026-08-21T17:07:27.400Z Google Research describes an AI system for prioritizing biomarkers from wearable data https://aiunderstanding.org/news/fda-opens-public-discussion-on-regulation-of-generative-ai-medical-devices https://aiunderstanding.org/api/news/image/fda-opens-public-discussion-on-regulation-of-generative-ai-medical-devices A primary-source healthcare technology illustration accompanying the FDA discussion of generative AI-enabled medical devices. AI Understanding en 2026-08-21T15:12:35.146Z FDA Opens Public Discussion on Regulation of Generative AI Medical Devices https://aiunderstanding.org/news/nvidia-previews-power-management-system-for-adding-ai-capacity-within-fixed-data-c https://aiunderstanding.org/api/news/image/nvidia-previews-power-management-system-for-adding-ai-capacity-within-fixed-data-c Liquid-cooled server racks and facility piping in a data-center equipment room. AI Understanding en 2026-08-21T15:04:43.163Z NVIDIA previews power-management system for adding AI capacity within fixed data-center budgets https://aiunderstanding.org/news/nvidia-reports-a-general-purpose-agent-system-solved-arc-agi-3-s-public-set https://aiunderstanding.org/api/news/image/nvidia-reports-a-general-purpose-agent-system-solved-arc-agi-3-s-public-set Rack-mounted GPU servers and dense fiber cables in a quiet high-performance computing data-center aisle at dawn. AI Understanding en 2026-08-21T13:08:36.757Z NVIDIA reports a general-purpose agent system solved ARC-AGI-3’s public set https://aiunderstanding.org/news/nvidia-says-ai-agent-security-should-sit-below-the-harness https://aiunderstanding.org/api/news/image/nvidia-says-ai-agent-security-should-sit-below-the-harness Empty secured data-center aisle with server racks, fiber cables and a mesh access partition AI Understanding en 2026-08-21T13:07:51.211Z NVIDIA says AI-agent security should sit below the harness https://aiunderstanding.org/news/google-deepmind-plans-staged-ai-research-in-eve-s-persistent-game-worlds https://aiunderstanding.org/api/news/image/google-deepmind-plans-staged-ai-research-in-eve-s-persistent-game-worlds An empty, unbranded game-testing room with a keyboard, mouse, desktop tower, server rack and small spacecraft model. AI Understanding en 2026-08-21T12:05:19.249Z Google DeepMind plans staged AI research in EVE’s persistent game worlds https://aiunderstanding.org/news/open-source-app-warns-about-nearby-smart-glasses-but-its-signals-are-only-a-rough https://aiunderstanding.org/api/news/image/open-source-app-warns-about-nearby-smart-glasses-but-its-signals-are-only-a-rough Unbranded camera-equipped smart glasses beside a face-down smartphone on a café table AI Understanding en 2026-08-21T11:10:18.266Z Open-source app warns about nearby smart glasses, but its signals are only a rough proxy https://aiunderstanding.org/news/google-threat-intelligence-reports-ai-assisted-vulnerability-research-and-autonomo https://aiunderstanding.org/api/news/image/google-threat-intelligence-reports-ai-assisted-vulnerability-research-and-autonomo A face-down Android phone beside a disassembled router, network cables and a locked server rack in a dim security laboratory. AI Understanding en 2026-08-21T09:23:34.643Z Google Threat Intelligence reports AI-assisted vulnerability research and autonomous malware https://aiunderstanding.org/news/preprint-reports-language-model-outputs-can-leak-secrets-held-in-context https://aiunderstanding.org/api/news/image/preprint-reports-language-model-outputs-can-leak-secrets-held-in-context Unoccupied university machine-learning server room with locked cabinets, fiber-optic cables and small indicator lights. AI Understanding en 2026-08-21T04:30:45.294Z Preprint reports language-model outputs can leak secrets held in context https://aiunderstanding.org/news/thinkingbox-benchmark-finds-ai-agents-struggle-to-reliably-complete-stateful-busin https://aiunderstanding.org/api/news/image/thinkingbox-benchmark-finds-ai-agents-struggle-to-reliably-complete-stateful-busin An empty office IT workspace with a closed laptop, network equipment and filing cabinet in early morning light, evoking the systems behind stateful business workflows. AI Understanding en 2026-08-21T04:21:50.258Z Thinkingbox benchmark finds AI agents struggle to reliably complete stateful business workflows https://aiunderstanding.org/news/preprint-reports-asymmetric-risks-that-llm-compression-metrics-can-miss https://aiunderstanding.org/api/news/image/preprint-reports-asymmetric-risks-that-llm-compression-metrics-can-miss An empty academic data-center aisle with compact server racks, storage modules, and neatly routed cables under cool dawn lighting AI Understanding en 2026-08-21T04:21:06.788Z Preprint reports asymmetric risks that LLM compression metrics can miss https://aiunderstanding.org/news/preprint-finds-irrelevant-text-can-shift-multimodal-model-judgments-predictably https://aiunderstanding.org/api/news/image/preprint-finds-irrelevant-text-can-shift-multimodal-model-judgments-predictably A quiet university computer-vision laboratory at late afternoon, with a black industrial camera aimed at two identical color calibration boards on a metal optical table and several blank white cards arranged beside them; no people are present. AI Understanding en 2026-08-21T04:10:27.978Z Preprint finds irrelevant text can shift multimodal model judgments predictably https://aiunderstanding.org/news/preprint-proposes-singularclip-to-preserve-learning-plasticity-in-changing-tasks https://aiunderstanding.org/api/news/image/preprint-proposes-singularclip-to-preserve-learning-plasticity-in-changing-tasks An empty machine-learning laboratory with matte-black server racks, a metal workstation, and loose network cables under cool evening lighting. AI Understanding en 2026-08-21T02:46:37.516Z Preprint proposes SingularClip to preserve learning plasticity in changing tasks https://aiunderstanding.org/news/preprint-proposes-a-probe-tested-gate-for-dynamic-ensembles-under-distribution-shi https://aiunderstanding.org/api/news/image/preprint-proposes-a-probe-tested-gate-for-dynamic-ensembles-under-distribution-shi Illustrative photograph of three unlabeled sensor boxes facing a grid of calibration tiles on a laboratory workbench. AI Understanding en 2026-08-21T02:43:47.240Z Preprint proposes a probe-tested gate for dynamic ensembles under distribution shift https://aiunderstanding.org/news/preprint-traces-how-llama-3-1-8b-models-numerical-sequence-structure https://aiunderstanding.org/api/news/image/preprint-traces-how-llama-3-1-8b-models-numerical-sequence-structure An empty university machine-learning laboratory at dawn with an unbranded server rack, patch cables and a blank whiteboard marked only with faint unlabeled lines and arrows. AI Understanding en 2026-08-21T02:41:57.277Z Preprint traces how Llama 3.1 8B models numerical sequence structure https://aiunderstanding.org/news/preprint-proposes-adaptive-feature-selection-for-personalized-fall-prevention https://aiunderstanding.org/api/news/image/preprint-proposes-adaptive-feature-selection-for-personalized-fall-prevention An empty physical-therapy room with a plain wearable activity band beside a wooden balance rail on rubber flooring. AI Understanding en 2026-08-21T02:39:20.674Z Preprint proposes adaptive feature selection for personalized fall prevention https://aiunderstanding.org/news/preprint-proposes-ecg-model-for-atrial-fibrillation-detection-across-lead-configur https://aiunderstanding.org/api/news/image/preprint-proposes-ecg-model-for-atrial-fibrillation-detection-across-lead-configur An empty hospital cardiology examination room with ECG lead wires and disposable electrodes arranged on a stainless-steel tray beside a wheeled recording cart. AI Understanding en 2026-08-21T02:35:53.234Z Preprint proposes ECG model for atrial-fibrillation detection across lead configurations https://aiunderstanding.org/news/preprint-proposes-physics-guided-neural-operator-for-wireless-radio-maps https://aiunderstanding.org/api/news/image/preprint-proposes-physics-guided-neural-operator-for-wireless-radio-maps An empty indoor radio-propagation test room with a ceiling-mounted wireless access point and a tripod-mounted antenna between partition walls. AI Understanding en 2026-08-21T02:33:45.964Z Preprint proposes physics-guided neural operator for wireless radio maps https://aiunderstanding.org/news/preprint-proposes-physics-constrained-generative-ai-for-global-tropical-cyclone-fo https://aiunderstanding.org/api/news/image/preprint-proposes-physics-constrained-generative-ai-for-global-tropical-cyclone-fo A representative coastal weather station beneath dense storm clouds over Okinawa, illustrating tropical-cyclone forecasting. AI Understanding en 2026-08-21T02:29:46.867Z Preprint proposes physics-constrained generative AI for global tropical-cyclone forecasts https://aiunderstanding.org/news/preprint-reports-llm-scheduler-cut-data-center-energy-use-32-and-waiting-time-30 https://aiunderstanding.org/api/news/image/preprint-reports-llm-scheduler-cut-data-center-energy-use-32-and-waiting-time-30 A quiet GPU server aisle in an unnamed data center, with black metal racks, fiber-optic cables, cooling vents, and small amber status lights. AI Understanding en 2026-08-21T02:25:31.553Z Preprint reports LLM scheduler cut data-center energy use 32% and waiting time 30% https://aiunderstanding.org/news/study-proposes-interaction-metric-for-explaining-llm-prompt-sensitivity https://aiunderstanding.org/api/news/image/study-proposes-interaction-metric-for-explaining-llm-prompt-sensitivity Blank index cards arranged in small groups beside a keyboard in an empty university machine-learning laboratory. AI Understanding en 2026-08-21T02:23:20.097Z Study proposes interaction metric for explaining LLM prompt sensitivity https://aiunderstanding.org/news/preprint-introduces-a-benchmark-and-replay-method-for-continual-reasoning-model-tr https://aiunderstanding.org/api/news/image/preprint-introduces-a-benchmark-and-replay-method-for-continual-reasoning-model-tr An empty university machine-learning laboratory with black server cabinets and cooling equipment under cool overhead lights at dawn. AI Understanding en 2026-08-21T02:17:51.986Z Preprint introduces a benchmark and replay method for continual reasoning-model training https://aiunderstanding.org/news/preprint-says-sequence-wide-normalization-can-bypass-a-convolutional-receptive-fie https://aiunderstanding.org/api/news/image/preprint-says-sequence-wide-normalization-can-bypass-a-convolutional-receptive-fie Illustrative computational-biology laboratory with colored sequence tiles beside an unbranded server rack. AI Understanding en 2026-08-21T02:17:04.500Z Preprint says sequence-wide normalization can bypass a convolutional receptive field https://aiunderstanding.org/news/arxiv-preprint-reports-related-slow-dynamics-in-transformer-and-mamba-blocks https://aiunderstanding.org/api/news/image/arxiv-preprint-reports-related-slow-dynamics-in-transformer-and-mamba-blocks Unbranded server racks and organized cable bundles in an empty university computing laboratory at dawn. AI Understanding en 2026-08-21T02:12:23.551Z ArXiv Preprint Reports Related Slow Dynamics in Transformer and Mamba Blocks https://aiunderstanding.org/news/preprint-reports-off-manifold-collapse-in-guided-protein-language-models https://aiunderstanding.org/api/news/image/preprint-reports-off-manifold-collapse-in-guided-protein-language-models An empty protein-engineering laboratory bench with unmarked glass vials, a stainless-steel tube rack, pipette tips, and blue nitrile gloves under cool morning window light. AI Understanding en 2026-08-21T02:08:40.707Z Preprint Reports Off-Manifold Collapse in Guided Protein Language Models https://aiunderstanding.org/news/study-finds-denoising-can-recover-more-text-from-noise-protected-embeddings https://aiunderstanding.org/api/news/image/study-finds-denoising-can-recover-more-text-from-noise-protected-embeddings A locked metal server rack with bundled fiber-optic cables beside a plain storage cabinet in a university data-center corridor, photographed before sunrise under cool overhead lights. AI Understanding en 2026-08-21T02:05:57.929Z Study Finds Denoising Can Recover More Text From Noise-Protected Embeddings https://aiunderstanding.org/news/preprint-proposes-a-test-for-structural-bias-in-llm-discussions-of-the-middle-east https://aiunderstanding.org/api/news/image/preprint-proposes-a-test-for-structural-bias-in-llm-discussions-of-the-middle-east An empty, unbranded computing laboratory in Abu Dhabi with closed laptops and blank evaluation cards on a metal table in early morning light. AI Understanding en 2026-08-21T01:54:20.407Z Preprint proposes a test for structural bias in LLM discussions of the Middle East https://aiunderstanding.org/news/preprint-reports-a-label-light-valence-direction-that-transfers-across-four-modali https://aiunderstanding.org/api/news/image/preprint-reports-a-label-light-valence-direction-that-transfers-across-four-modali An empty university neuroscience laboratory bench with an EEG cap, microphone and abstract image card under mixed evening light. AI Understanding en 2026-08-21T01:50:53.076Z Preprint reports a label-light valence direction that transfers across four modalities https://aiunderstanding.org/news/preprint-describes-a-multi-agent-ai-framework-for-analyzing-chinese-herbal-formula https://aiunderstanding.org/api/news/image/preprint-describes-a-multi-agent-ai-framework-for-analyzing-chinese-herbal-formula An empty traditional Chinese medicine research laboratory with unlabeled botanical samples and laboratory tools arranged on a wooden bench. AI Understanding en 2026-08-21T01:49:43.214Z Preprint describes a multi-agent AI framework for analyzing Chinese herbal formulas https://aiunderstanding.org/news/ph-d-dissertation-examines-backdoor-attacks-in-language-and-vision-language-models https://aiunderstanding.org/api/news/image/ph-d-dissertation-examines-backdoor-attacks-in-language-and-vision-language-models Unbranded server cabinet in an empty university computer laboratory, illustrating research into AI model backdoors. AI Understanding en 2026-08-21T01:46:54.584Z Ph.D. dissertation examines backdoor attacks in language and vision-language models https://aiunderstanding.org/news/preprint-reports-higher-llm-candidate-ratings-for-prestigious-institutions https://aiunderstanding.org/api/news/image/preprint-reports-higher-llm-candidate-ratings-for-prestigious-institutions Empty university review room in Guayaquil with unmarked application folders and a closed academic journal on a wooden table. AI Understanding en 2026-08-21T01:38:55.066Z Preprint reports higher LLM candidate ratings for prestigious institutions https://aiunderstanding.org/news/study-finds-inference-context-can-change-llm-outputs-in-medical-allocation-scenari https://aiunderstanding.org/api/news/image/study-finds-inference-context-can-change-llm-outputs-in-medical-allocation-scenari An empty clinical simulation room with two hospital beds and a stainless-steel medical supply cart, illustrating research on language-model behavior in medical resource allocation. AI Understanding en 2026-08-21T01:34:03.368Z Study finds inference context can change LLM outputs in medical allocation scenarios https://aiunderstanding.org/news/preprint-reports-ai-decoding-of-natural-sentences-from-non-invasive-brain-recordin https://aiunderstanding.org/api/news/image/preprint-reports-ai-decoding-of-natural-sentences-from-non-invasive-brain-recordin Empty university MEG research room with a whole-head sensor helmet above a reclined chair and a plain keyboard on a side table. AI Understanding en 2026-08-21T01:30:41.925Z Preprint reports AI decoding of natural sentences from non-invasive brain recordings https://aiunderstanding.org/news/study-maps-46-language-models-built-for-portuguese https://aiunderstanding.org/api/news/image/study-maps-46-language-models-built-for-portuguese Unlabeled research cards and computing equipment arranged on a table in a university library research room. AI Understanding en 2026-08-21T01:28:51.659Z Study maps 46 language models built for Portuguese https://aiunderstanding.org/news/interpretability-study-links-qwen3-4b-s-overconfident-answers-to-a-certainty-biase https://aiunderstanding.org/api/news/image/interpretability-study-links-qwen3-4b-s-overconfident-answers-to-a-certainty-biase A close view of a GPU server rack and organized cables in an empty university machine-learning laboratory during late afternoon. AI Understanding en 2026-08-21T01:17:17.959Z Interpretability study links Qwen3-4B’s overconfident answers to a certainty-biased mechanism https://aiunderstanding.org/news/the-deontic-gap-large-language-models-and-the-modal-language-of-obligation https://aiunderstanding.org/api/news/image/the-deontic-gap-large-language-models-and-the-modal-language-of-obligation A photograph of a library with books on shelves and a few people sitting at tables reading. AI Understanding en 2026-08-21T01:12:51.316Z The Deontic Gap: Large Language Models and the Modal Language of Obligation https://aiunderstanding.org/news/paper-proposes-obscuring-refusal-signals-to-resist-abliteration-attacks https://aiunderstanding.org/api/news/image/paper-proposes-obscuring-refusal-signals-to-resist-abliteration-attacks An unbranded server rack and circuit board in a machine-learning research lab. AI Understanding en 2026-08-21T01:05:06.220Z Paper proposes obscuring refusal signals to resist abliteration attacks https://aiunderstanding.org/news/paper-reports-a-temporal-method-for-detecting-hallucinations-at-the-token-level https://aiunderstanding.org/api/news/image/paper-reports-a-temporal-method-for-detecting-hallucinations-at-the-token-level An empty machine-learning laboratory with a server rack and unmarked colored tokens arranged in a line on a wooden workbench. AI Understanding en 2026-08-21T01:01:23.263Z Paper reports a temporal method for detecting hallucinations at the token level https://aiunderstanding.org/news/entity-tracking-emerges-at-410-million-parameters-exceeds-humans-across-naturalist https://aiunderstanding.org/api/news/image/entity-tracking-emerges-at-410-million-parameters-exceeds-humans-across-naturalist A photograph of a bookshelf with books on language and linguistics. AI Understanding en 2026-08-21T00:59:36.560Z Entity tracking emerges at 410 million parameters, exceeds humans across naturalistic narratives https://aiunderstanding.org/news/paper-reports-compiler-guided-search-improves-lean-theorem-proving-efficiency https://aiunderstanding.org/api/news/image/paper-reports-compiler-guided-search-improves-lean-theorem-proving-efficiency A quiet university computer laboratory at dusk with a metal server rack, coiled black cables, cooling vents, and small indicator lights, with no people present. AI Understanding en 2026-08-21T00:58:03.815Z Paper reports compiler-guided search improves Lean theorem-proving efficiency https://aiunderstanding.org/news/benchmark-separates-geometry-solving-from-diagram-construction https://aiunderstanding.org/api/news/image/benchmark-separates-geometry-solving-from-diagram-construction An empty university mathematics classroom with a geometric construction drawn on a whiteboard. AI Understanding en 2026-08-21T00:54:24.923Z Benchmark Separates Geometry Solving From Diagram Construction https://aiunderstanding.org/news/alignment-is-all-you-need-instruction-free-training-for-general-audio-language-mod https://aiunderstanding.org/api/news/image/alignment-is-all-you-need-instruction-free-training-for-general-audio-language-mod A photograph of a computer screen displaying a waveform of an audio signal, with a caption indicating that the model learns from (audio, response) pairs without explicit task instructions. AI Understanding en 2026-08-21T00:50:47.740Z Alignment Is All You Need: Instruction-Free Training for General Audio-Language Models https://aiunderstanding.org/news/are-llms-safe-beyond-text-do-emojis-expose-gaps-in-safety-evaluation https://aiunderstanding.org/api/news/image/are-llms-safe-beyond-text-do-emojis-expose-gaps-in-safety-evaluation A photograph of a computer screen displaying a large language model (LLM) interface, with a variety of emojis and text inputs. AI Understanding en 2026-08-21T00:46:23.801Z Are LLMs Safe Beyond Text: Do Emojis Expose Gaps in Safety Evaluation https://aiunderstanding.org/news/bridging-search-and-crm-productionizing-ai-product-research-agents-for-customer-re https://aiunderstanding.org/api/news/image/bridging-search-and-crm-productionizing-ai-product-research-agents-for-customer-re A photograph of a smartphone screen displaying a WhatsApp conversation with a personalized recommendation for a mobile product. AI Understanding en 2026-08-21T00:41:06.199Z Bridging Search and CRM: Productionizing AI Product Research Agents for Customer Re-Engagement https://aiunderstanding.org/news/new-benchmark-shows-vietnam-s-exam-rubric-can-change-how-language-models-rank https://aiunderstanding.org/api/news/image/new-benchmark-shows-vietnam-s-exam-rubric-can-change-how-language-models-rank An empty Vietnamese high-school examination classroom with rows of wooden desks and closed, unmarked answer booklets under mixed daylight and fluorescent light. AI Understanding en 2026-08-21T00:39:49.754Z New benchmark shows Vietnam’s exam rubric can change how language models rank https://aiunderstanding.org/news/netflix-paper-outlines-a-lifecycle-for-llm-judges-evaluating-recommendation-explan https://aiunderstanding.org/api/news/image/netflix-paper-outlines-a-lifecycle-for-llm-judges-evaluating-recommendation-explan Unlabeled server racks with blue status lights in a streaming-service data center, illustrating infrastructure for large-scale recommendation evaluation. AI Understanding en 2026-08-21T00:29:33.607Z Netflix paper outlines a lifecycle for LLM judges evaluating recommendation explanations https://aiunderstanding.org/news/survey-frames-self-evolving-ai-agents-as-dynamic-graph-transformations https://aiunderstanding.org/api/news/image/survey-frames-self-evolving-ai-agents-as-dynamic-graph-transformations An empty university computer-science laboratory with colored node tiles and connecting lines arranged on a corkboard beside an unbranded metal workstation. AI Understanding en 2026-08-21T00:21:18.529Z Survey frames self-evolving AI agents as dynamic graph transformations https://aiunderstanding.org/news/facet-proposes-an-environment-grounded-method-for-training-terminal-agents https://aiunderstanding.org/api/news/image/facet-proposes-an-environment-grounded-method-for-training-terminal-agents An empty university computer-science laboratory with a workstation keyboard, dark monitor, network cables and a small unmarked server chassis. AI Understanding en 2026-08-21T00:18:44.725Z FACET proposes an environment-grounded method for training terminal agents https://aiunderstanding.org/news/sesse-proposes-structured-decomposition-to-make-llm-judging-more-auditable https://aiunderstanding.org/api/news/image/sesse-proposes-structured-decomposition-to-make-llm-judging-more-auditable A quiet university computer-science laboratory with blank evaluation cards and colored sorting tokens beside an unbranded server rack. AI Understanding en 2026-08-21T00:16:02.958Z SESSE proposes structured decomposition to make LLM judging more auditable https://aiunderstanding.org/news/ibm-team-reports-ai-written-adapters-bring-thousands-of-hugging-face-models-to-spy https://aiunderstanding.org/api/news/image/ibm-team-reports-ai-written-adapters-bring-thousands-of-hugging-face-models-to-spy An unbranded AI accelerator card and circuit boards arranged on an anti-static workbench in a hardware test laboratory. AI Understanding en 2026-08-21T00:06:13.845Z IBM team reports AI-written adapters bring thousands of Hugging Face models to Spyre https://aiunderstanding.org/news/apple-researchers-report-iterative-pseudo-labeling-gains-for-mandarin-english-spee https://aiunderstanding.org/api/news/image/apple-researchers-report-iterative-pseudo-labeling-gains-for-mandarin-english-spee An empty university speech laboratory with microphones, acoustic panels and audio equipment AI Understanding en 2026-08-20T23:50:15.165Z Apple researchers report iterative pseudo-labeling gains for Mandarin-English speech recognition https://aiunderstanding.org/news/google-to-let-users-tune-discover-with-natural-language-requests https://aiunderstanding.org/api/news/image/google-to-let-users-tune-discover-with-natural-language-requests An unbranded smartphone beside blank newsprint and preference cards on a public library reading table in early morning light. AI Understanding en 2026-08-20T21:53:54.627Z Google to let users tune Discover with natural-language requests https://aiunderstanding.org/news/aws-adds-cross-region-access-to-openai-gpt-5-6-models-on-bedrock https://aiunderstanding.org/api/news/image/aws-adds-cross-region-access-to-openai-gpt-5-6-models-on-bedrock An empty data-center aisle in Northern Virginia with black server racks, overhead cable trays, and bundled fiber-optic cables under cool white lights. AI Understanding en 2026-08-20T21:49:46.902Z AWS adds cross-Region access to OpenAI GPT-5.6 models on Bedrock https://aiunderstanding.org/news/google-adds-publisher-preference-labels-and-citation-badges-to-ai-search https://aiunderstanding.org/api/news/image/google-adds-publisher-preference-labels-and-citation-badges-to-ai-search An empty public-library reading table with an unbranded rack of newspapers in early morning light. AI Understanding en 2026-08-20T19:22:49.714Z Google adds publisher-preference labels and citation badges to AI Search https://aiunderstanding.org/news/mit-researchers-develop-computational-method-to-search-for-greener-ammonia-catalys https://aiunderstanding.org/api/news/image/mit-researchers-develop-computational-method-to-search-for-greener-ammonia-catalys An ammonia production plant with pipes, storage tanks, and cooling equipment in early-morning light, illustrating the industrial process that greener catalyst research seeks to improve. AI Understanding en 2026-08-20T19:07:50.827Z MIT researchers develop computational method to search for greener ammonia catalysts https://aiunderstanding.org/news/nvidia-backed-tests-explore-flexible-ai-factory-power-use-during-grid-stress https://aiunderstanding.org/api/news/image/nvidia-backed-tests-explore-flexible-ai-factory-power-use-during-grid-stress Rows of server racks, power cabling and cooling equipment in an empty London AI-factory hall. AI Understanding en 2026-08-20T17:49:50.396Z NVIDIA-Backed Tests Explore Flexible AI-Factory Power Use During Grid Stress https://aiunderstanding.org/news/ramp-launches-router-a-multi-model-gateway-that-automatically-balances-cost-and-pe https://aiunderstanding.org/api/news/image/ramp-launches-router-a-multi-model-gateway-that-automatically-balances-cost-and-pe Unmarked server racks and fiber-optic cables in a quiet U.S. data-center equipment room at dawn AI Understanding en 2026-08-20T16:50:09.424Z Ramp launches Router, a multi-model gateway that automatically balances cost and performance https://aiunderstanding.org/news/aws-expands-agentcore-policy-authoring-for-natural-language-controls-on-ai-agents https://aiunderstanding.org/api/news/image/aws-expands-agentcore-policy-authoring-for-natural-language-controls-on-ai-agents Unoccupied cloud data-center aisle with server racks, cabling, and access-control hardware under cool white lights. AI Understanding en 2026-08-20T16:35:00.475Z AWS expands AgentCore policy authoring for natural-language controls on AI agents https://aiunderstanding.org/news/microsoft-expands-access-to-skala-1-1-for-predictive-chemistry-simulations https://aiunderstanding.org/api/news/image/microsoft-expands-access-to-skala-1-1-for-predictive-chemistry-simulations An empty computational chemistry laboratory with physical molecular models beside a matte-black scientific computing chassis on a stainless-steel workbench. AI Understanding en 2026-08-20T16:05:32.698Z Microsoft expands access to Skala 1.1 for predictive chemistry simulations https://aiunderstanding.org/news/code-agents-lose-reliability-when-code-is-rewritten-without-changing-its-meaning-s https://aiunderstanding.org/api/news/image/code-agents-lose-reliability-when-code-is-rewritten-without-changing-its-meaning-s An empty university software-testing laboratory at dawn, with closed laptops, Ethernet cables and an unmarked server rack on workbenches. AI Understanding en 2026-08-20T16:01:14.206Z Code agents lose reliability when code is rewritten without changing its meaning, study finds https://aiunderstanding.org/news/enhanced-fuzzy-logic-model-for-power-transformer-fault-diagnosis-using-ieee-key-ga https://aiunderstanding.org/api/news/image/enhanced-fuzzy-logic-model-for-power-transformer-fault-diagnosis-using-ieee-key-ga A photograph of a power transformer in a substation, with a large metal tank and a series of electrical connections. AI Understanding en 2026-08-20T15:35:11.277Z Enhanced Fuzzy Logic Model for Power Transformer Fault Diagnosis Using IEEE Key Gas Method Improvements https://aiunderstanding.org/news/new-preprint-reports-gains-from-looped-language-models-in-multi-step-tool-calling https://aiunderstanding.org/api/news/image/new-preprint-reports-gains-from-looped-language-models-in-multi-step-tool-calling Unlabeled network appliances linked by colored cables on a workbench in a university computing laboratory AI Understanding en 2026-08-20T15:17:41.919Z New preprint reports gains from looped language models in multi-step tool calling https://aiunderstanding.org/news/adversarial-review-tests-structured-disagreement-for-agentic-code-review https://aiunderstanding.org/api/news/image/adversarial-review-tests-structured-disagreement-for-agentic-code-review An empty software testing room with three unoccupied workstations, blank monitors, printed circuit boards, and a glass-walled code laboratory in evening light AI Understanding en 2026-08-20T15:09:20.092Z Adversarial Review tests structured disagreement for agentic code review https://aiunderstanding.org/news/arxiv-study-reports-large-roman-urdu-hate-speech-gains-from-lora-adaptation https://aiunderstanding.org/api/news/image/arxiv-study-reports-large-roman-urdu-hate-speech-gains-from-lora-adaptation Unmarked annotation cards, a keyboard and research equipment on a desk in a university language laboratory, illustrating Roman Urdu hate-speech detection research. AI Understanding en 2026-08-20T15:04:57.900Z ArXiv study reports large Roman Urdu hate-speech gains from LoRA adaptation https://aiunderstanding.org/news/preliminary-fraudbench-test-finds-banking-agents-vulnerable-to-adaptive-fraud https://aiunderstanding.org/api/news/image/preliminary-fraudbench-test-finds-banking-agents-vulnerable-to-adaptive-fraud An empty generic bank customer-service counter with a telephone, keypad and locked cash drawer under evening lights. AI Understanding en 2026-08-20T15:00:20.953Z Preliminary FraudBench test finds banking agents vulnerable to adaptive fraud https://aiunderstanding.org/news/apple-researchers-report-scaling-law-for-training-models-with-scarce-data https://aiunderstanding.org/api/news/image/apple-researchers-report-scaling-law-for-training-models-with-scarce-data Unoccupied data-center equipment room with server cabinets and coiled fiber cables in early morning light AI Understanding en 2026-08-20T14:52:22.814Z Apple researchers report scaling law for training models with scarce data https://aiunderstanding.org/news/apple-researchers-propose-lexical-substitutions-to-improve-multilingual-model-trai https://aiunderstanding.org/api/news/image/apple-researchers-propose-lexical-substitutions-to-improve-multilingual-model-trai An unoccupied computational-linguistics laboratory with server racks and two open bilingual dictionaries on a wooden worktable, their pages too distant to read. AI Understanding en 2026-08-20T14:50:35.850Z Apple Researchers Propose Lexical Substitutions to Improve Multilingual Model Training https://aiunderstanding.org/news/position-paper-calls-for-certification-before-ai-agents-make-market-decisions https://aiunderstanding.org/api/news/image/position-paper-calls-for-certification-before-ai-agents-make-market-decisions An empty university economics computer laboratory with paired workstations, calculators and blank cards arranged for a market simulation. AI Understanding en 2026-08-20T14:47:58.247Z Position paper calls for certification before AI agents make market decisions https://aiunderstanding.org/news/systematic-review-maps-the-growing-use-of-large-language-models-in-mental-health https://aiunderstanding.org/api/news/image/systematic-review-maps-the-growing-use-of-large-language-models-in-mental-health An empty mental-health clinic consultation room with two chairs, a plain table, and an unbranded closed laptop beside a small audio microphone. AI Understanding en 2026-08-20T14:43:30.013Z Systematic review maps the growing use of large language models in mental health https://aiunderstanding.org/news/model-cards-alone-may-not-govern-open-weight-foundation-models-position-paper-argu https://aiunderstanding.org/api/news/image/model-cards-alone-may-not-govern-open-weight-foundation-models-position-paper-argu An empty university computing laboratory with unbranded folders, blank documentation sheets, and a small server chassis arranged on a wooden table. AI Understanding en 2026-08-20T14:40:52.278Z Model Cards Alone May Not Govern Open-Weight Foundation Models, Position Paper Argues https://aiunderstanding.org/news/a-proposed-metric-would-measure-how-difficult-game-worlds-are-to-predict https://aiunderstanding.org/api/news/image/a-proposed-metric-would-measure-how-difficult-game-worlds-are-to-predict An empty university AI research laboratory at dawn with an unbranded game controller beside a tabletop grid of colored wooden tiles and a small stack of blank cards, lit by cool window light and a single warm desk lamp. AI Understanding en 2026-08-20T14:38:40.974Z A proposed metric would measure how difficult game worlds are to predict https://aiunderstanding.org/news/fm-bench-tests-whether-ai-agents-can-manage-a-football-club-for-20-years https://aiunderstanding.org/api/news/image/fm-bench-tests-whether-ai-agents-can-manage-a-football-club-for-20-years An empty football club management office with a tactics board, football, and blank folders. AI Understanding en 2026-08-20T04:15:32.105Z FM-Bench tests whether AI agents can manage a football club for 20 years https://aiunderstanding.org/news/finrca-bench-finds-financial-ai-diagnosis-depends-heavily-on-evidence-retrieval https://aiunderstanding.org/api/news/image/finrca-bench-finds-financial-ai-diagnosis-depends-heavily-on-evidence-retrieval An empty finance records room at dawn with metal shelving, sealed cardboard archive boxes, closed folders, a calculator, and an unplugged desk lamp under cool window light. AI Understanding en 2026-08-20T04:10:26.695Z FinRCA-Bench finds financial AI diagnosis depends heavily on evidence retrieval https://aiunderstanding.org/news/abra-paper-maps-compute-and-data-tradeoffs-in-diffusion-image-training https://aiunderstanding.org/api/news/image/abra-paper-maps-compute-and-data-tradeoffs-in-diffusion-image-training An empty university data-center aisle with unbranded GPU server racks, fiber cables, and cooling vents under early-morning white lights. AI Understanding en 2026-08-20T03:31:15.754Z Abra paper maps compute and data tradeoffs in diffusion image training https://aiunderstanding.org/news/delta2gamma-reports-92-4-accuracy-using-eeg-to-detect-alzheimer-s-disease https://aiunderstanding.org/api/news/image/delta2gamma-reports-92-4-accuracy-using-eeg-to-detect-alzheimer-s-disease An empty clinical EEG examination room with an EEG cap, electrode cables, conductive gel containers, and an unlabeled amplifier arranged on a stainless-steel tray beside an examination chair. AI Understanding en 2026-08-20T03:11:57.123Z Delta2Gamma reports 92.4% accuracy using EEG to detect Alzheimer’s disease https://aiunderstanding.org/news/agentic-esopt-proposes-lower-memory-fine-tuning-for-long-horizon-ai-agents https://aiunderstanding.org/api/news/image/agentic-esopt-proposes-lower-memory-fine-tuning-for-long-horizon-ai-agents Unoccupied university machine-learning laboratory with compact GPU workstations arranged on a metal rack AI Understanding en 2026-08-20T03:03:58.200Z Agentic ESOpt proposes lower-memory fine-tuning for long-horizon AI agents https://aiunderstanding.org/news/paper-proposes-cse-to-evaluate-irregular-time-series-forecasts-beyond-mse https://aiunderstanding.org/api/news/image/paper-proposes-cse-to-evaluate-irregular-time-series-forecasts-beyond-mse Unoccupied university laboratory bench with unbranded sensor modules and a data-logging unit arranged for irregular measurements. AI Understanding en 2026-08-20T03:02:16.577Z Paper proposes CSE to evaluate irregular time-series forecasts beyond MSE https://aiunderstanding.org/news/study-proposes-transfer-aware-curriculum-sampling-for-language-model-training https://aiunderstanding.org/api/news/image/study-proposes-transfer-aware-curriculum-sampling-for-language-model-training An empty university machine-learning laboratory with a server rack, fiber cables, and unmarked hardware used for an illustrative story about adaptive language-model training. AI Understanding en 2026-08-20T02:55:20.047Z Study proposes transfer-aware curriculum sampling for language-model training https://aiunderstanding.org/news/co-rl-paper-reports-label-free-reasoning-gains-from-diverse-model-cohorts https://aiunderstanding.org/api/news/image/co-rl-paper-reports-label-free-reasoning-gains-from-diverse-model-cohorts An empty university machine-learning laboratory with unbranded server racks and separated workstations connected by fiber cables. AI Understanding en 2026-08-20T02:52:47.317Z Co-RL paper reports label-free reasoning gains from diverse model cohorts https://aiunderstanding.org/news/a-proposed-fine-tuning-method-allocates-training-effort-across-related-tasks https://aiunderstanding.org/api/news/image/a-proposed-fine-tuning-method-allocates-training-effort-across-related-tasks An unbranded machine-learning laboratory worktable beside a whiteboard with unlabeled task-allocation diagrams, illustrating research on specialized fine-tuning. AI Understanding en 2026-08-20T02:46:11.807Z A proposed fine-tuning method allocates training effort across related tasks https://aiunderstanding.org/news/sure-ridge-proposes-a-faster-way-to-infer-causal-graphs-from-limited-data https://aiunderstanding.org/api/news/image/sure-ridge-proposes-a-faster-way-to-infer-causal-graphs-from-limited-data An empty university statistics laboratory with a chalkboard showing an unlabeled directed graph of circles and arrows beside a desk with a calculator and plain paper. AI Understanding en 2026-08-20T02:26:16.517Z SURE-Ridge proposes a faster way to infer causal graphs from limited data https://aiunderstanding.org/news/j-miner-paper-reports-executable-decision-rules-extracted-from-language-model-clas https://aiunderstanding.org/api/news/image/j-miner-paper-reports-executable-decision-rules-extracted-from-language-model-clas A rack server, blank cards, and laboratory tools arranged on a wooden table in a university machine-learning laboratory. AI Understanding en 2026-08-20T02:16:11.736Z J-Miner paper reports executable decision rules extracted from language-model classifiers https://aiunderstanding.org/news/paper-tests-letting-ai-agents-switch-specialized-lora-adapters-mid-task https://aiunderstanding.org/api/news/image/paper-tests-letting-ai-agents-switch-specialized-lora-adapters-mid-task Two unbranded GPU server towers and blank module cases in an empty university machine-learning laboratory. AI Understanding en 2026-08-20T02:10:27.155Z Paper tests letting AI agents switch specialized LoRA adapters mid-task https://aiunderstanding.org/news/multisigbert-combines-medical-text-and-patient-timelines-for-cancer-survival-predi https://aiunderstanding.org/api/news/image/multisigbert-combines-medical-text-and-patient-timelines-for-cancer-survival-predi An empty oncology records room with paper folders and clinical equipment, representing research that combines medical reports and structured patient data. AI Understanding en 2026-08-20T02:06:35.072Z MultiSigBERT combines medical text and patient timelines for cancer survival prediction https://aiunderstanding.org/news/dow-ke-proposes-direct-optimization-for-editing-knowledge-across-ai-model-layers https://aiunderstanding.org/api/news/image/dow-ke-proposes-direct-optimization-for-editing-knowledge-across-ai-model-layers An empty university machine-learning laboratory with an unbranded GPU server rack and a workbench of blank research materials. AI Understanding en 2026-08-20T02:02:28.913Z DOW-KE proposes direct optimization for editing knowledge across AI model layers https://aiunderstanding.org/news/study-finds-public-eeg-encoders-can-transfer-adversarial-attacks-to-private-models https://aiunderstanding.org/api/news/image/study-finds-public-eeg-encoders-can-transfer-adversarial-attacks-to-private-models An empty university EEG recording room with a white electrode cap, loose sensor wires, and laboratory recording equipment arranged on a stainless-steel cart. AI Understanding en 2026-08-20T02:00:54.284Z Study Finds Public EEG Encoders Can Transfer Adversarial Attacks to Private Models https://aiunderstanding.org/news/paper-proposes-mr-dec-to-model-hospital-stays-day-by-day-for-30-day-readmission-pr https://aiunderstanding.org/api/news/image/paper-proposes-mr-dec-to-model-hospital-stays-day-by-day-for-30-day-readmission-pr An empty hospital radiology archive room with a blank lightbox displaying a de-identified chest radiograph beside closed paper folders and metal storage cabinets. AI Understanding en 2026-08-20T01:58:58.647Z Paper proposes Mr.Dec to model hospital stays day by day for 30-day readmission prediction https://aiunderstanding.org/news/a-leakage-controlled-benchmark-tests-ai-classification-of-sensitive-documents https://aiunderstanding.org/api/news/image/a-leakage-controlled-benchmark-tests-ai-classification-of-sensitive-documents A closed archive box and a laptop on a wooden table in a quiet records room, with all screens and labels turned away from view AI Understanding en 2026-08-20T01:57:48.206Z A leakage-controlled benchmark tests AI classification of sensitive documents https://aiunderstanding.org/news/data-dpo-paper-proposes-target-model-aware-selection-of-fine-tuning-data https://aiunderstanding.org/api/news/image/data-dpo-paper-proposes-target-model-aware-selection-of-fine-tuning-data Unmarked sample trays beside a server rack in an academic machine-learning laboratory. AI Understanding en 2026-08-20T01:55:04.164Z Data-DPO paper proposes target-model-aware selection of fine-tuning data https://aiunderstanding.org/news/stripe-agrees-to-acquire-openrouter-to-expand-ai-model-routing https://aiunderstanding.org/api/news/image/stripe-agrees-to-acquire-openrouter-to-expand-ai-model-routing Unbranded server rack with dense network cables in a softly lit data-center aisle. AI Understanding en 2026-08-19T23:36:23.090Z Stripe agrees to acquire OpenRouter to expand AI model routing