<?xml version="1.0" encoding="UTF-8"?><rss version="2.0"><channel><title>ResearchNode · AI news</title><link>https://researchnode.dev/news</link><description>AI-assisted briefings with original sources and automated source-support checks.</description><language>en</language><item><guid isPermaLink="true">https://researchnode.dev/news/what-balanced-accuracy-cannot-identify-in-neural-signal-decoding-under-test-time-adap-79bd88b1</guid><title>Balanced accuracy can miss a dropped class in test-time adaptation, arXiv paper argues</title><link>https://researchnode.dev/news/what-balanced-accuracy-cannot-identify-in-neural-signal-decoding-under-test-time-adap-79bd88b1</link><description>A new arXiv study of neural-signal decoding shows aggregate scores can stay flat while a model silently stops predicting one class for some subjects.</description><pubDate>Tue, 17 Nov 2026 06:12:34 GMT</pubDate></item><item><guid isPermaLink="true">https://researchnode.dev/news/fend-far-end-detection-in-outbound-voice-ai-calls-f1e55ed6</guid><title>FEND classifies who answers outbound Voice AI calls: human, voicemail, IVR or another AI</title><link>https://researchnode.dev/news/fend-far-end-detection-in-outbound-voice-ai-calls-f1e55ed6</link><description>Researchers behind the FEND framework say standard answering-machine detection is too coarse for AI agents making calls, and propose per-class detectors that label the far end within 20 seconds.</description><pubDate>Tue, 17 Nov 2026 06:11:19 GMT</pubDate></item><item><guid isPermaLink="true">https://researchnode.dev/news/poetry-resurrection-from-llms-d82f591e</guid><title>Researchers test whether AI can write an infinite poem</title><link>https://researchnode.dev/news/poetry-resurrection-from-llms-d82f591e</link><description>An arXiv paper reports two experiments using couplets to probe how far large language models can extend a poem, finding interesting output that still trails human creativity.</description><pubDate>Tue, 17 Nov 2026 06:10:06 GMT</pubDate></item><item><guid isPermaLink="true">https://researchnode.dev/news/adversarial-tracking-attacks-on-image-sensors-for-object-tracking-in-autonomous-vehic-ade88da2</guid><title>Adversarial image attacks sharply degrade vehicle tracking in simulated autonomous driving tests</title><link>https://researchnode.dev/news/adversarial-tracking-attacks-on-image-sensors-for-object-tracking-in-autonomous-vehic-ade88da2</link><description>A new arXiv paper reports that adversarial perturbations to camera images can collapse the accuracy of tracking systems built on YOLO detectors, with YOLOv3-based trackers hit hardest in CARLA simulations.</description><pubDate>Tue, 17 Nov 2026 06:08:51 GMT</pubDate></item><item><guid isPermaLink="true">https://researchnode.dev/news/gt-feng-general-theory-of-factual-evaluation-and-normative-grounding-in-language-mode-c5eef725</guid><title>GT-FENG framework ties language model fact-checking to applicable rules</title><link>https://researchnode.dev/news/gt-feng-general-theory-of-factual-evaluation-and-normative-grounding-in-language-mode-c5eef725</link><description>Researchers behind the arXiv paper GT-FENG propose evaluating model answers against the rules that apply in a given context, arguing correctness in law and policy depends on jurisdiction, date and definitions.</description><pubDate>Tue, 17 Nov 2026 06:06:29 GMT</pubDate></item><item><guid isPermaLink="true">https://researchnode.dev/news/learning-word-level-explanations-of-text-reuse-in-latin-literature-with-little-annota-dd99950b</guid><title>Pointer network explains Latin text reuse word by word with 200 annotated pairs</title><link>https://researchnode.dev/news/learning-word-level-explanations-of-text-reuse-in-latin-literature-with-little-annota-dd99950b</link><description>Researchers describe a model that aligns each word of a reusing Latin passage to its source and labels the change, trained with far less annotation than before.</description><pubDate>Tue, 17 Nov 2026 06:05:14 GMT</pubDate></item><item><guid isPermaLink="true">https://researchnode.dev/news/ai-reviews-are-biased-towards-the-appendix-2b956330</guid><title>Study finds GPT and Claude review weaknesses cluster in paper appendices</title><link>https://researchnode.dev/news/ai-reviews-are-biased-towards-the-appendix-2b956330</link><description>Researchers analyzing 2,000 papers report that leading LLMs draw roughly half or more of their review weaknesses from appendices, while human reviewers draw under a tenth from there.</description><pubDate>Tue, 17 Nov 2026 06:04:14 GMT</pubDate></item><item><guid isPermaLink="true">https://researchnode.dev/news/interactive-program-induction-with-language-model-priors-63d40bee</guid><title>Interactive program induction method uses language model priors to query users efficiently</title><link>https://researchnode.dev/news/interactive-program-induction-with-language-model-priors-63d40bee</link><description>A new arXiv paper frames program induction as a Bayesian learning problem, deriving a greedy query algorithm that reportedly beats a language-model-only agent on CodeARC.</description><pubDate>Tue, 17 Nov 2026 06:02:58 GMT</pubDate></item><item><guid isPermaLink="true">https://researchnode.dev/news/evaluating-human-sensitivity-to-gender-biased-ai-generated-content-through-human-asse-7ea9dcfc</guid><title>Study finds AI attribution does not change how readers judge gender-biased statements</title><link>https://researchnode.dev/news/evaluating-human-sensitivity-to-gender-biased-ai-generated-content-through-human-asse-7ea9dcfc</link><description>A Spanish-language experiment using Google autocomplete statements found that telling participants content was AI-generated did not alter their agreement or reaction times.</description><pubDate>Tue, 17 Nov 2026 06:01:42 GMT</pubDate></item><item><guid isPermaLink="true">https://researchnode.dev/news/two-clustering-regime-of-token-dynamics-in-causal-attention-ce51402f</guid><title>Mathematicians partially prove when attention tokens split into two clusters</title><link>https://researchnode.dev/news/two-clustering-regime-of-token-dynamics-in-causal-attention-ce51402f</link><description>An arXiv paper resolves part of a 2024 conjecture about causal attention, showing token dynamics form two clusters when a key eigenvalue is positive, and generally not otherwise.</description><pubDate>Tue, 17 Nov 2026 06:00:43 GMT</pubDate></item><item><guid isPermaLink="true">https://researchnode.dev/news/what-llm-agents-see-and-what-users-say-837ba15b</guid><title>Study ties agent write safety to how lookups reveal competing records</title><link>https://researchnode.dev/news/what-llm-agents-see-and-what-users-say-837ba15b</link><description>Researchers apply linguistic presupposition theory to LLM agents, finding that whether a tool lists, counts, or hides a second record strongly changes how often agents act on false assumptions.</description><pubDate>Tue, 17 Nov 2026 05:59:32 GMT</pubDate></item><item><guid isPermaLink="true">https://researchnode.dev/news/outfitterbench-evaluating-llm-agents-on-long-horizon-resource-allocation-and-commitme-8b1b0b44</guid><title>OutfitterBench tests LLM agents on long-horizon resource allocation</title><link>https://researchnode.dev/news/outfitterbench-evaluating-llm-agents-on-long-horizon-resource-allocation-and-commitme-8b1b0b44</link><description>A new benchmark asks language-model agents to run a simulated rental business for 200 days, where their own commitments shape future options. Most tested models trail a simple baseline.</description><pubDate>Tue, 17 Nov 2026 05:56:55 GMT</pubDate></item><item><guid isPermaLink="true">https://researchnode.dev/news/vision-as-a-tool-visual-harness-for-multimodal-agentic-coding-3b22b352</guid><title>Vision-as-a-Tool paradigm treats multimodal models as callable utilities for coding agents</title><link>https://researchnode.dev/news/vision-as-a-tool-visual-harness-for-multimodal-agentic-coding-3b22b352</link><description>The VHarness framework lets text-only coding agents offload visual reasoning to external multimodal models, with reported gains in image, SVG and 3D generation tasks.</description><pubDate>Tue, 17 Nov 2026 05:55:36 GMT</pubDate></item><item><guid isPermaLink="true">https://researchnode.dev/news/litarc-bench-disentangling-text-specific-reader-response-dynamics-from-position-and-l-44a2dd8d</guid><title>LitArc-Bench tests whether AI models track literary reader response beyond position and word cues</title><link>https://researchnode.dev/news/litarc-bench-disentangling-text-specific-reader-response-dynamics-from-position-and-l-44a2dd8d</link><description>A new benchmark separates narrative position, affective vocabulary, and text-specific response dynamics in model predictions of human emotional reactions to literature.</description><pubDate>Tue, 17 Nov 2026 05:53:07 GMT</pubDate></item><item><guid isPermaLink="true">https://researchnode.dev/news/towards-explainable-sleep-stage-classification-using-single-channel-electroencephalog-ac0b6d85</guid><title>SleepXAI study finds Integrated Gradients most faithful for explaining single-channel EEG sleep staging</title><link>https://researchnode.dev/news/towards-explainable-sleep-stage-classification-using-single-channel-electroencephalog-ac0b6d85</link><description>An arXiv paper extends the SleepXAI framework to compare five explainable AI methods, reporting that Integrated Gradients gave the most faithful attributions across two datasets.</description><pubDate>Tue, 17 Nov 2026 05:51:48 GMT</pubDate></item><item><guid isPermaLink="true">https://researchnode.dev/news/cross-benchmark-transfer-from-rl-on-agentic-coding-tasks-4a37f09b</guid><title>RL training on 1,700 coding tasks lifts Kimi K2.7 Code across six benchmarks</title><link>https://researchnode.dev/news/cross-benchmark-transfer-from-rl-on-agentic-coding-tasks-4a37f09b</link><description>The authors report that reinforcement learning alone, applied to expert-built agentic coding tasks, improved pass@1 on every external benchmark they tested, including sets released after training data was collected.</description><pubDate>Tue, 17 Nov 2026 05:49:26 GMT</pubDate></item><item><guid isPermaLink="true">https://researchnode.dev/news/cpm-multiagent-a-cpm-grounded-appraisal-framework-for-simulating-dynamic-patient-emot-9c6e4c59</guid><title>CPM-MultiAgent simulates how patient emotions shift during healthcare training dialogues</title><link>https://researchnode.dev/news/cpm-multiagent-a-cpm-grounded-appraisal-framework-for-simulating-dynamic-patient-emot-9c6e4c59</link><description>Researchers propose a multi-agent framework that models a simulated patient&#39;s emotional evolution turn by turn, aiming to give medical trainees richer feedback than static emotion portrayals.</description><pubDate>Tue, 17 Nov 2026 05:48:01 GMT</pubDate></item><item><guid isPermaLink="true">https://researchnode.dev/news/boostlora-growing-effective-rank-by-boosting-adapters-0e11d047</guid><title>BoostLoRA trains stacked low-rank adapters to lift effective rank</title><link>https://researchnode.dev/news/boostlora-growing-effective-rank-by-boosting-adapters-0e11d047</link><description>A new arXiv paper describes a gradient-boosting approach to parameter-efficient fine-tuning that iteratively merges tiny adapters, reporting gains on Qwen2.5-3B math and code benchmarks with no added inference cost.</description><pubDate>Tue, 17 Nov 2026 05:46:46 GMT</pubDate></item><item><guid isPermaLink="true">https://researchnode.dev/news/abflownet-optimizing-antibody-antigen-binding-energy-via-diffusion-gflownet-fusion-9b7b827a</guid><title>AbFlowNet fuses diffusion and GFlowNet training to design antibody binding regions</title><link>https://researchnode.dev/news/abflownet-optimizing-antibody-antigen-binding-energy-via-diffusion-gflownet-fusion-9b7b827a</link><description>The arXiv paper proposes jointly training a generative model on diffusion losses and binding energy, replacing costly reinforcement learning pipelines for antibody CDR design.</description><pubDate>Tue, 17 Nov 2026 05:43:56 GMT</pubDate></item><item><guid isPermaLink="true">https://researchnode.dev/news/efficiently-aligning-draft-models-via-parameter-and-data-efficient-adaptation-2e303bf0</guid><title>EDA adapts speculative decoding draft models to fine-tuned LLMs without full retraining</title><link>https://researchnode.dev/news/efficiently-aligning-draft-models-via-parameter-and-data-efficient-adaptation-2e303bf0</link><description>Researchers propose Efficient Draft Adaptation, a framework that realigns draft models with fine-tuned target models using shared and private components plus regenerated training data, cutting retraining costs.</description><pubDate>Tue, 17 Nov 2026 05:42:18 GMT</pubDate></item><item><guid isPermaLink="true">https://researchnode.dev/news/hidden-not-deleted-how-networks-suppress-entangled-features-ca16112d</guid><title>Silent outputs can hide features rather than forget them, arXiv study finds</title><link>https://researchnode.dev/news/hidden-not-deleted-how-networks-suppress-entangled-features-ca16112d</link><description>A small-model study traces how a neural network can suppress a feature while retaining it, and why a quiet output alone does not prove the feature is gone.</description><pubDate>Tue, 17 Nov 2026 05:41:24 GMT</pubDate></item><item><guid isPermaLink="true">https://researchnode.dev/news/scrapegraphai-100k-dataset-for-schema-constrained-llm-generation-e5324f83</guid><title>ScrapeGraphAI-100k dataset pairs real web pages with schema-constrained LLM extraction data</title><link>https://researchnode.dev/news/scrapegraphai-100k-dataset-for-schema-constrained-llm-generation-e5324f83</link><description>Researchers released 93,695 schema-constrained extraction events from opt-out telemetry, aiming to fill a gap left by small, synthetic or text-only public datasets.</description><pubDate>Tue, 17 Nov 2026 05:40:08 GMT</pubDate></item><item><guid isPermaLink="true">https://researchnode.dev/news/larch-learned-query-optimization-for-semantic-predicates-5b81eb2b</guid><title>Larch optimizes semantic filters in AI SQL queries to cut token costs</title><link>https://researchnode.dev/news/larch-learned-query-optimization-for-semantic-predicates-5b81eb2b</link><description>A research paper introduces Larch, a framework that reorders and optimizes semantic filter execution in database queries, reporting lower token costs than existing techniques across tested text and image workloads.</description><pubDate>Tue, 17 Nov 2026 05:38:51 GMT</pubDate></item><item><guid isPermaLink="true">https://researchnode.dev/news/brain2vec-a-self-supervised-3d-foundation-model-for-structural-brain-mri-98f16eef</guid><title>Brain2Vec: open-source 3D foundation model for structural brain MRI released</title><link>https://researchnode.dev/news/brain2vec-a-self-supervised-3d-foundation-model-for-structural-brain-mri-98f16eef</link><description>Researchers pre-trained a self-supervised model on 74,425 T1-weighted MRI scans, reporting gains in brain age prediction and benchmark rankings, with code and models openly available.</description><pubDate>Tue, 17 Nov 2026 05:37:37 GMT</pubDate></item><item><guid isPermaLink="true">https://researchnode.dev/news/when-does-on-policy-interaction-help-representational-tradeoffs-in-value-based-imitat-9b69785f</guid><title>Interactive imitation learning eases what a model must represent, arXiv study finds</title><link>https://researchnode.dev/news/when-does-on-policy-interaction-help-representational-tradeoffs-in-value-based-imitat-9b69785f</link><description>A new arXiv paper argues that querying the expert along the learner&#39;s own trajectories shifts the burden from copying the expert&#39;s policy to matching its value function.</description><pubDate>Tue, 17 Nov 2026 05:36:19 GMT</pubDate></item><item><guid isPermaLink="true">https://researchnode.dev/news/graph-your-own-prompt-3951e696</guid><title>Graph Consistency Regularization uses a model&#39;s own predictions to shape its features</title><link>https://researchnode.dev/news/graph-your-own-prompt-3951e696</link><description>A new arXiv paper proposes GCR, a training framework that aligns a network&#39;s internal feature relationships with its predicted class structure, aiming for cleaner representations without architectural changes.</description><pubDate>Tue, 17 Nov 2026 05:35:02 GMT</pubDate></item><item><guid isPermaLink="true">https://researchnode.dev/news/wavelet-flow-matching-for-multi-scale-physics-emulation-ce38a29c</guid><title>Wavelet Flow Matching targets stable generative emulation of multi-scale physics</title><link>https://researchnode.dev/news/wavelet-flow-matching-for-multi-scale-physics-emulation-ce38a29c</link><description>An arXiv paper proposes a generative emulator that works directly in wavelet space, aiming to avoid the cost of latent autoencoders while keeping fine-scale detail in long rollouts.</description><pubDate>Tue, 17 Nov 2026 05:33:46 GMT</pubDate></item><item><guid isPermaLink="true">https://researchnode.dev/news/from-data-to-program-fast-amp-direct-generative-program-inference-from-empirical-data-76a98cfc</guid><title>PRODiGI infers executable generative programs from data in a single pass</title><link>https://researchnode.dev/news/from-data-to-program-fast-amp-direct-generative-program-inference-from-empirical-data-76a98cfc</link><description>Researchers present PRODiGI, a pretrained model that turns sample data into an explicit, runnable generative program without dataset-specific fitting, and report accuracy and speed gains over competing pretrained models.</description><pubDate>Tue, 17 Nov 2026 05:32:30 GMT</pubDate></item><item><guid isPermaLink="true">https://researchnode.dev/news/vision-language-models-lag-human-performance-on-physical-dynamics-and-intent-reasonin-7ca35963</guid><title>New EscherVerse benchmark finds vision-language models far behind humans on goal-driven spatial reasoning</title><link>https://researchnode.dev/news/vision-language-models-lag-human-performance-on-physical-dynamics-and-intent-reasonin-7ca35963</link><description>A research team&#39;s Teleo-Spatial Intelligence benchmark tests whether AI models can connect physical dynamics with human intentions, and 27 leading models still trail human annotators by a wide margin.</description><pubDate>Tue, 17 Nov 2026 05:31:14 GMT</pubDate></item><item><guid isPermaLink="true">https://researchnode.dev/news/comparative-analysis-of-3d-convolutional-and-2-5d-slice-conditioned-u-net-architectur-ac874aab</guid><title>Study finds no overall winner between 3D and 2.5D U-Nets for diffusion-based MRI super-resolution</title><link>https://researchnode.dev/news/comparative-analysis-of-3d-convolutional-and-2-5d-slice-conditioned-u-net-architectur-ac874aab</link><description>An arXiv abstract reports that two U-Net backbones for an elucidated diffusion MRI super-resolution model perform comparably, with each leading under different test conditions.</description><pubDate>Tue, 17 Nov 2026 05:29:58 GMT</pubDate></item><item><guid isPermaLink="true">https://researchnode.dev/news/deltabox-millisecond-scale-sandbox-checkpoint-and-rollback-for-stateful-ai-agents-956a0290</guid><title>DeltaBox brings millisecond checkpoint and rollback to AI agent sandboxes</title><link>https://researchnode.dev/news/deltabox-millisecond-scale-sandbox-checkpoint-and-rollback-for-stateful-ai-agents-956a0290</link><description>A research paper describes an OS-level design that checkpoints and restores agent sandboxes in milliseconds by copying only state changes instead of full snapshots.</description><pubDate>Tue, 17 Nov 2026 05:28:42 GMT</pubDate></item><item><guid isPermaLink="true">https://researchnode.dev/news/hackworld-evaluating-computer-use-agents-on-exploiting-web-application-vulnerabilitie-7899e8da</guid><title>HackWorld benchmark finds computer-use agents exploit under 12% of web vulnerabilities</title><link>https://researchnode.dev/news/hackworld-evaluating-computer-use-agents-on-exploiting-web-application-vulnerabilitie-7899e8da</link><description>A new arXiv paper introduces HackWorld, a framework testing whether computer-use agents can find and exploit real web application flaws through visual interaction, and reports weak results.</description><pubDate>Tue, 17 Nov 2026 05:27:25 GMT</pubDate></item><item><guid isPermaLink="true">https://researchnode.dev/news/can-llms-improve-their-own-harness-a-controlled-study-on-arc-agi-3-b285d909</guid><title>Study finds LLMs build their own task harnesses, but self-revision gains stay small</title><link>https://researchnode.dev/news/can-llms-improve-their-own-harness-a-controlled-study-on-arc-agi-3-b285d909</link><description>A controlled arXiv study had frontier models construct and twice revise harnesses for ARC-AGI-3. Autonomous revisions helped little on held-out games, though selecting among snapshots by development score did better.</description><pubDate>Tue, 17 Nov 2026 05:26:09 GMT</pubDate></item><item><guid isPermaLink="true">https://researchnode.dev/news/courteous-anticipation-improving-long-lived-task-planning-in-persistent-shared-enviro-1b0720f3</guid><title>Courteous anticipatory planning cuts long-run robot task costs in shared spaces</title><link>https://researchnode.dev/news/courteous-anticipation-improving-long-lived-task-planning-in-persistent-shared-enviro-1b0720f3</link><description>A new arXiv paper proposes a planner that weighs how today&#39;s robot actions affect future tasks for every robot in a shared environment, reporting lower total cost than myopic and selfish baselines.</description><pubDate>Tue, 17 Nov 2026 05:24:53 GMT</pubDate></item><item><guid isPermaLink="true">https://researchnode.dev/news/timesjev-thinking-fast-on-time-series-with-system-1-sufficiency-bb5e3051</guid><title>TimesJev compresses time series for instant LLM-style reasoning</title><link>https://researchnode.dev/news/timesjev-thinking-fast-on-time-series-with-system-1-sufficiency-bb5e3051</link><description>The arXiv paper proposes a System 1 model that compresses numeric series into a compact representation, cutting inference time while the authors report higher accuracy than long chain-of-thought reasoning models.</description><pubDate>Tue, 17 Nov 2026 05:23:37 GMT</pubDate></item><item><guid isPermaLink="true">https://researchnode.dev/news/neat-neighborhood-guided-efficient-autoregressive-set-transformer-for-3d-molecular-ge-a9b6409a</guid><title>NEAT reorders autoregressive 3D molecule generation around atom neighborhoods</title><link>https://researchnode.dev/news/neat-neighborhood-guided-efficient-autoregressive-set-transformer-for-3d-molecular-ge-a9b6409a</link><description>The authors of the NEAT paper say their neighborhood-guided training removes the need for canonical atom orderings while matching or beating diffusion and flow baselines on quality and speed.</description><pubDate>Tue, 17 Nov 2026 05:22:21 GMT</pubDate></item><item><guid isPermaLink="true">https://researchnode.dev/news/generative-machine-learning-for-protein-design-48467a6a</guid><title>Review maps 59 generative AI approaches for protein design</title><link>https://researchnode.dev/news/generative-machine-learning-for-protein-design-48467a6a</link><description>An arXiv review organizes generative methods for protein design into six task categories and weighs how well current evaluation methods and limitations support reliable design.</description><pubDate>Tue, 17 Nov 2026 05:21:04 GMT</pubDate></item><item><guid isPermaLink="true">https://researchnode.dev/news/input-pathways-shape-few-shot-not-zero-shot-binding-in-tiny-transformers-a-fully-enum-6447170d</guid><title>Readability, not input pathway, drives few-shot binding in tiny transformers, study finds</title><link>https://researchnode.dev/news/input-pathways-shape-few-shot-not-zero-shot-binding-in-tiny-transformers-a-fully-enum-6447170d</link><description>A fully enumerable study of 6-10K-parameter transformers reports that zero-shot compositional binding fails on all tested input routes, while few-shot success tracks how readable the input code is.</description><pubDate>Tue, 17 Nov 2026 05:20:08 GMT</pubDate></item><item><guid isPermaLink="true">https://researchnode.dev/news/exposing-and-mitigating-temporal-attack-in-deepfake-video-detection-c139c15b</guid><title>SpInShield targets temporal shortcuts that let attacks evade deepfake video detectors</title><link>https://researchnode.dev/news/exposing-and-mitigating-temporal-attack-in-deepfake-video-detection-c139c15b</link><description>Researchers behind the SpInShield framework argue that spatiotemporal deepfake detectors rely on fragile spectral cues, and report large AUC gains under simulated spectral attacks.</description><pubDate>Tue, 17 Nov 2026 05:18:53 GMT</pubDate></item><item><guid isPermaLink="true">https://researchnode.dev/news/qval-evaluating-dense-supervision-signals-for-long-horizon-llm-agents-f77d4fe4</guid><title>QVal testbed scores how well dense reward signals rank agent actions</title><link>https://researchnode.dev/news/qval-evaluating-dense-supervision-signals-for-long-horizon-llm-agents-f77d4fe4</link><description>A training-free benchmark measures whether intermediate-step scoring methods order candidate actions the same way a reference policy&#39;s returns do, offering a cheaper check before full training runs.</description><pubDate>Tue, 17 Nov 2026 05:17:38 GMT</pubDate></item></channel></rss>