https://aiunderstanding.org/news/systematic-review-of-55-studies-finds-autism-machine-learning-still-leans-on-super https://aiunderstanding.org/api/news/image/systematic-review-of-55-studies-finds-autism-machine-learning-still-leans-on-super An empty developmental assessment room in a university clinic, with wooden play materials on a low table and a one-way observation window AI Understanding en 2026-08-23T16:14:15.568Z Systematic review of 55 studies finds autism machine learning still leans on supervised models https://aiunderstanding.org/news/preprint-ties-model-uncertainty-to-the-label-tree-reporting-about-half-the-calibra https://aiunderstanding.org/api/news/image/preprint-ties-model-uncertainty-to-the-label-tree-reporting-about-half-the-calibra Primary-source image accompanying Preprint ties model uncertainty to the label tree, reporting about half the calibration error AI Understanding en 2026-08-23T15:55:03.992Z Preprint ties model uncertainty to the label tree, reporting about half the calibration error https://aiunderstanding.org/news/preprint-reports-steering-representation-geometry-makes-brain-model-alignment-more https://aiunderstanding.org/api/news/image/preprint-reports-steering-representation-geometry-makes-brain-model-alignment-more Rack-mounted computing servers in a university research machine room, with bundled cables and perforated steel doors under overhead lighting AI Understanding en 2026-08-23T15:44:44.259Z Preprint reports steering representation geometry makes brain–model alignment more two-way https://aiunderstanding.org/news/preprint-proposes-batched-parallel-decoding-to-cut-training-time-for-compact-image https://aiunderstanding.org/api/news/image/preprint-proposes-batched-parallel-decoding-to-cut-training-time-for-compact-image Primary-source image accompanying Preprint proposes batched parallel decoding to cut training time for compact image-generation models AI Understanding en 2026-08-23T03:26:21.386Z Preprint proposes batched parallel decoding to cut training time for compact image-generation models https://aiunderstanding.org/news/philosophy-paper-argues-machine-learning-should-borrow-its-standards-of-proof-from https://aiunderstanding.org/api/news/image/philosophy-paper-argues-machine-learning-should-borrow-its-standards-of-proof-from Empty seminar room at a medical school in late afternoon, with a bare oak table, a wiped chalkboard and a glass cabinet of old apothecary bottles AI Understanding en 2026-08-23T03:22:43.672Z Philosophy paper argues machine learning should borrow its standards of proof from clinical translation https://aiunderstanding.org/news/preprint-casts-interpretability-tools-as-one-measurement-problem-tests-it-on-gpt-2 https://aiunderstanding.org/api/news/image/preprint-casts-interpretability-tools-as-one-measurement-problem-tests-it-on-gpt-2 Primary-source image accompanying Preprint casts interpretability tools as one measurement problem, tests it on GPT-2 and Qwen AI Understanding en 2026-08-23T02:48:24.089Z Preprint casts interpretability tools as one measurement problem, tests it on GPT-2 and Qwen https://aiunderstanding.org/news/preprint-proposes-a-continuous-alzheimer-s-severity-score-learned-from-repeated-br https://aiunderstanding.org/api/news/image/preprint-proposes-a-continuous-alzheimer-s-severity-score-learned-from-repeated-br Primary-source image accompanying Preprint proposes a continuous Alzheimer's severity score learned from repeated brain scans AI Understanding en 2026-08-23T02:35:34.604Z Preprint proposes a continuous Alzheimer's severity score learned from repeated brain scans https://aiunderstanding.org/news/preprint-reports-physics-informed-network-gains-came-from-one-pairing-broke-when-s https://aiunderstanding.org/api/news/image/preprint-reports-physics-informed-network-gains-came-from-one-pairing-broke-when-s Primary-source image accompanying Preprint reports physics-informed network gains came from one pairing, broke when stacked AI Understanding en 2026-08-23T02:15:43.904Z Preprint reports physics-informed network gains came from one pairing, broke when stacked https://aiunderstanding.org/news/paper-says-a-recommender-trained-only-on-synthetic-clickstreams-leads-zero-shot-be https://aiunderstanding.org/api/news/image/paper-says-a-recommender-trained-only-on-synthetic-clickstreams-leads-zero-shot-be Primary-source image accompanying Paper says a recommender trained only on synthetic clickstreams leads zero-shot benchmarks AI Understanding en 2026-08-23T02:09:20.146Z Paper says a recommender trained only on synthetic clickstreams leads zero-shot benchmarks https://aiunderstanding.org/news/deltaml-bench-finds-agent-scaffolding-changes-success-on-machine-learning-research https://aiunderstanding.org/api/news/image/deltaml-bench-finds-agent-scaffolding-changes-success-on-machine-learning-research Primary-source image accompanying DeltaML-Bench finds agent scaffolding changes success on machine-learning research tasks AI Understanding en 2026-08-23T01:36:16.880Z DeltaML-Bench finds agent scaffolding changes success on machine-learning research tasks https://aiunderstanding.org/news/preprint-proposes-answer-level-trust-checks-for-physical-vision-language-model-pre https://aiunderstanding.org/api/news/image/preprint-proposes-answer-level-trust-checks-for-physical-vision-language-model-pre Primary-source image accompanying Preprint proposes answer-level trust checks for physical vision-language model predictions AI Understanding en 2026-08-23T01:34:19.394Z Preprint proposes answer-level trust checks for physical vision-language model predictions https://aiunderstanding.org/news/preprint-audit-finds-common-credit-signals-fail-to-identify-causally-important-ste https://aiunderstanding.org/api/news/image/preprint-audit-finds-common-credit-signals-fail-to-identify-causally-important-ste Blank index cards arranged in branching rows on a wooden table in an empty university laboratory. AI Understanding en 2026-08-23T01:28:34.492Z Preprint audit finds common credit signals fail to identify causally important steps in LLM agents https://aiunderstanding.org/news/preprint-proposes-adaptive-safety-shields-for-reinforcement-learning-agents https://aiunderstanding.org/api/news/image/preprint-proposes-adaptive-safety-shields-for-reinforcement-learning-agents Primary-source image accompanying Preprint proposes adaptive safety shields for reinforcement-learning agents AI Understanding en 2026-08-23T01:26:23.148Z Preprint proposes adaptive safety shields for reinforcement-learning agents https://aiunderstanding.org/news/paper-outlines-assurance-path-for-an-onboard-ml-helicopter-weight-estimator https://aiunderstanding.org/api/news/image/paper-outlines-assurance-path-for-an-onboard-ml-helicopter-weight-estimator An unmarked helicopter resting on industrial weighing platforms inside a maintenance hangar. AI Understanding en 2026-08-23T01:25:08.315Z Paper outlines assurance path for an onboard ML helicopter-weight estimator https://aiunderstanding.org/news/deltamomentum-paper-proposes-direction-aware-optimizer-updates-for-neural-network https://aiunderstanding.org/api/news/image/deltamomentum-paper-proposes-direction-aware-optimizer-updates-for-neural-network Unbranded GPU servers and cooling hardware in an academic machine-learning lab equipment room AI Understanding en 2026-08-23T01:23:03.334Z DeltaMomentum paper proposes direction-aware optimizer updates for neural-network training https://aiunderstanding.org/news/transformer-study-estimates-days-before-severe-copd-flare-ups-from-home-ventilator https://aiunderstanding.org/api/news/image/transformer-study-estimates-days-before-severe-copd-flare-ups-from-home-ventilator Primary-source image accompanying Transformer study estimates days before severe COPD flare-ups from home-ventilator data AI Understanding en 2026-08-23T01:20:45.909Z Transformer study estimates days before severe COPD flare-ups from home-ventilator data https://aiunderstanding.org/news/preprint-proposes-a-two-hemisphere-architecture-for-continual-learning https://aiunderstanding.org/api/news/image/preprint-proposes-a-two-hemisphere-architecture-for-continual-learning Primary-source image accompanying Preprint proposes a two-hemisphere architecture for continual learning AI Understanding en 2026-08-23T01:03:10.278Z Preprint proposes a two-hemisphere architecture for continual learning https://aiunderstanding.org/news/paper-proposes-using-an-llm-to-generate-tabular-anomaly-detectors-from-normal-data https://aiunderstanding.org/api/news/image/paper-proposes-using-an-llm-to-generate-tabular-anomaly-detectors-from-normal-data Primary-source image accompanying Paper proposes using an LLM to generate tabular anomaly detectors from normal data AI Understanding en 2026-08-23T00:53:47.511Z Paper proposes using an LLM to generate tabular anomaly detectors from normal data https://aiunderstanding.org/news/eventtime-paper-proposes-ai-method-to-estimate-market-losses-after-cybersecurity-d https://aiunderstanding.org/api/news/image/eventtime-paper-proposes-ai-method-to-estimate-market-losses-after-cybersecurity-d Primary-source image accompanying EventTime paper proposes AI method to estimate market losses after cybersecurity disclosures AI Understanding en 2026-08-23T00:48:56.749Z EventTime paper proposes AI method to estimate market losses after cybersecurity disclosures https://aiunderstanding.org/news/preprint-proposes-a-low-cost-way-to-improve-confidence-estimates-for-black-box-llm https://aiunderstanding.org/api/news/image/preprint-proposes-a-low-cost-way-to-improve-confidence-estimates-for-black-box-llm Unbranded server cabinets and fiber-optic cables in a university AI research data-center room. AI Understanding en 2026-08-23T00:46:08.030Z Preprint proposes a low-cost way to improve confidence estimates for black-box LLMs https://aiunderstanding.org/news/fleetsieve-paper-proposes-targeted-profiling-for-slo-aware-llm-fleets https://aiunderstanding.org/api/news/image/fleetsieve-paper-proposes-targeted-profiling-for-slo-aware-llm-fleets Unbranded GPU server racks in a data-center aisle, illustrating LLM serving-fleet profiling AI Understanding en 2026-08-23T00:42:30.897Z FleetSieve paper proposes targeted profiling for SLO-aware LLM fleets https://aiunderstanding.org/news/sage-xgboost-paper-reports-stronger-landslide-and-wildfire-mapping-with-scarce-dat https://aiunderstanding.org/api/news/image/sage-xgboost-paper-reports-stronger-landslide-and-wildfire-mapping-with-scarce-dat Steep forested hillside above a dry valley, with exposed soil, rocks and a narrow drainage channel. AI Understanding en 2026-08-23T00:40:07.426Z SAGE-XGBoost paper reports stronger landslide and wildfire mapping with scarce data https://aiunderstanding.org/news/flashprefill-v2-paper-reports-large-long-context-serving-speedups-with-block-spars https://aiunderstanding.org/api/news/image/flashprefill-v2-paper-reports-large-long-context-serving-speedups-with-block-spars An empty server aisle with dense accelerator racks and organized cooling and network cabling in a large inference data center. AI Understanding en 2026-08-23T00:36:52.086Z FlashPrefill V2 paper reports large long-context serving speedups with block-sparse attention https://aiunderstanding.org/news/swe-bench-science-benchmark-finds-coding-agents-struggle-with-scientific-software https://aiunderstanding.org/api/news/image/swe-bench-science-benchmark-finds-coding-agents-struggle-with-scientific-software An empty university computational-science laboratory with unbranded computing and scientific equipment on a stainless-steel bench AI Understanding en 2026-08-23T00:25:45.366Z SWE-bench Science benchmark finds coding agents struggle with scientific software https://aiunderstanding.org/news/arxiv-paper-proposes-milestone-based-training-for-long-horizon-llm-agents https://aiunderstanding.org/api/news/image/arxiv-paper-proposes-milestone-based-training-for-long-horizon-llm-agents Primary-source image accompanying ArXiv paper proposes milestone-based training for long-horizon LLM agents AI Understanding en 2026-08-23T00:22:55.799Z ArXiv paper proposes milestone-based training for long-horizon LLM agents https://aiunderstanding.org/news/preprint-tests-whether-llm-agents-know-when-to-remember-verify-or-ask https://aiunderstanding.org/api/news/image/preprint-tests-whether-llm-agents-know-when-to-remember-verify-or-ask An empty university AI evaluation laboratory with an unbranded server rack and transparent trays of blank test cards on a stainless-steel workbench AI Understanding en 2026-08-23T00:20:34.047Z Preprint tests whether LLM agents know when to remember, verify or ask https://aiunderstanding.org/news/auditing-cross-lingual-fairness-in-language-model-watermarking https://aiunderstanding.org/api/news/image/auditing-cross-lingual-fairness-in-language-model-watermarking Primary-source image accompanying Auditing Cross-Lingual Fairness in Language Model Watermarking AI Understanding en 2026-08-23T00:18:50.899Z Auditing Cross-Lingual Fairness in Language Model Watermarking https://aiunderstanding.org/news/stanford-ai-index-finds-ai-policy-expanding-as-sovereignty-and-investment-diverge https://aiunderstanding.org/api/news/image/stanford-ai-index-finds-ai-policy-expanding-as-sovereignty-and-investment-diverge Primary-source image accompanying Stanford AI Index finds AI policy expanding as sovereignty and investment diverge AI Understanding en 2026-08-22T18:33:42.692Z Stanford AI Index finds AI policy expanding as sovereignty and investment diverge https://aiunderstanding.org/news/together-ai-benchmark-glm-5-3-trails-gpt-5-6-sol-on-the-first-try-wins-on-retries https://aiunderstanding.org/api/news/image/together-ai-benchmark-glm-5-3-trails-gpt-5-6-sol-on-the-first-try-wins-on-retries Rows of dark server cabinets filled with accelerator chassis in a data-hall aisle, with fiber cables running up to overhead ladder trays. AI Understanding en 2026-08-22T05:53:24.418Z Together AI benchmark: GLM-5.3 trails GPT-5.6 Sol on the first try, wins on retries at half the price https://aiunderstanding.org/news/preprint-proposes-a-locally-tokenized-ai-model-for-robust-time-series-watermarking https://aiunderstanding.org/api/news/image/preprint-proposes-a-locally-tokenized-ai-model-for-robust-time-series-watermarking Primary-source image accompanying Preprint proposes a locally tokenized AI model for robust time-series watermarking AI Understanding en 2026-08-22T00:59:12.480Z Preprint proposes a locally tokenized AI model for robust time-series watermarking https://aiunderstanding.org/news/natural-language-code-retrieval-for-1c-enterprise-an-open-benchmark-and-efficient https://aiunderstanding.org/api/news/image/natural-language-code-retrieval-for-1c-enterprise-an-open-benchmark-and-efficient A photograph of a computer screen displaying the 1C:Enterprise ecosystem, with code snippets and query-code pairs visible. AI Understanding en 2026-08-22T00:41:43.223Z Natural Language Code Retrieval for 1C:Enterprise: An Open Benchmark and Efficient Bi-Encoder https://aiunderstanding.org/news/time-series-retrieval-for-grounding-multimodal-language-models-in-remaining-useful https://aiunderstanding.org/api/news/image/time-series-retrieval-for-grounding-multimodal-language-models-in-remaining-useful Primary-source image accompanying Time-Series Retrieval for Grounding Multimodal Language Models in Remaining Useful Life AI Understanding en 2026-08-22T00:27:06.251Z Time-Series Retrieval for Grounding Multimodal Language Models in Remaining Useful Life https://aiunderstanding.org/news/nepali-english-preprint-text-only-ai-matched-multimodal-model-on-out-of-context-mi https://aiunderstanding.org/api/news/image/nepali-english-preprint-text-only-ai-matched-multimodal-model-on-out-of-context-mi Primary-source image accompanying Nepali-English preprint: text-only AI matched multimodal model on out-of-context misinformation benchmark AI Understanding en 2026-08-22T00:22:07.217Z Nepali-English preprint: text-only AI matched multimodal model on out-of-context misinformation benchmark https://aiunderstanding.org/news/preprint-finds-financial-ai-confidence-can-fail-when-text-changes https://aiunderstanding.org/api/news/image/preprint-finds-financial-ai-confidence-can-fail-when-text-changes An empty records-processing room with three separate stacks of unmarked paper on metal trays, representing financial filings, news material and social-media text being handled under different data conditions. AI Understanding en 2026-08-22T00:18:25.916Z Preprint finds financial AI confidence can fail when text changes https://aiunderstanding.org/news/can-conversational-ai-loosen-us-versus-them-boundaries https://aiunderstanding.org/api/news/image/can-conversational-ai-loosen-us-versus-them-boundaries A photograph of a person sitting in a room with a computer screen displaying a conversation with a large language model. AI Understanding en 2026-08-22T00:16:21.844Z Can Conversational AI loosen Us-Versus-Them Boundaries? https://aiunderstanding.org/news/preprint-finds-periodic-subject-changes-raise-judged-surprise-in-base-language-mod https://aiunderstanding.org/api/news/image/preprint-finds-periodic-subject-changes-raise-judged-surprise-in-base-language-mod Unbranded server cabinets and repeating groups of blank cards in an empty machine-learning laboratory AI Understanding en 2026-08-22T00:15:40.937Z Preprint finds periodic subject changes raise judged surprise in base language models https://aiunderstanding.org/news/preprint-reports-declining-diversity-in-llm-creative-outputs-over-three-years https://aiunderstanding.org/api/news/image/preprint-reports-declining-diversity-in-llm-creative-outputs-over-three-years Primary-source image accompanying Preprint reports declining diversity in LLM creative outputs over three years AI Understanding en 2026-08-22T00:11:42.314Z Preprint reports declining diversity in LLM creative outputs over three years https://aiunderstanding.org/news/paper-reports-a-fine-tuning-method-for-long-context-ai-with-sparse-attention https://aiunderstanding.org/api/news/image/paper-reports-a-fine-tuning-method-for-long-context-ai-with-sparse-attention An unbranded GPU server in a university research-computing lab, with exposed cooling hardware and coiled cables under cool dawn light. AI Understanding en 2026-08-22T00:09:16.651Z Paper reports a fine-tuning method for long-context AI with sparse attention https://aiunderstanding.org/news/google-cloud-details-codemender-ai-agent-for-code-vulnerability-remediation https://aiunderstanding.org/api/news/image/google-cloud-details-codemender-ai-agent-for-code-vulnerability-remediation Primary-source image accompanying Google Cloud Details CodeMender AI Agent for Code Vulnerability Remediation AI Understanding en 2026-08-21T21:58:16.651Z Google Cloud Details CodeMender AI Agent for Code Vulnerability Remediation