Research

Jaron Lanier argues that the term 'artificial intelligence' is misleading; large language models are statistical mashups of human creations, not independent minds. He advocates viewing AI as a tool, not a creature, and promotes data dignity and transparency to manage technological risks.

A new cyberattack called 'token torching' targets AI systems by exhausting their tokens through malicious prompts, potentially draining company resources and serving as a distraction for other attacks.

Venv-manager is a Go-based CLI tool for managing Python virtual environments, designed for both humans and AI agents. It offers a file watcher that auto-installs missing imports, an MCP server for agentic control, sandboxed ephemeral execution, and comprehensive venv lifecycle commands, addressing common venv sprawl issues.

Inertia-1 is a unified motion foundation model that tackles fragmentation in motion datasets. It uses large-scale self-supervised learning on wrist accelerometer data to create representations that generalize across body placements and sensor types, and reveals key design choices for real-world performance.

Tech professionals are feeling squeezed by AI. Burnout is up, optimism is down, survey found. One solution is saying 'no' deliberately.

A founder automated his support inbox using Claude Code and MCP servers: the agent classifies, investigates, and drafts replies, but never sends them. The post details the setup, key rules (no guessing, drafts only), and real-world results over 48 hours.

A study by Meta's Oversight Board finds that leading AI systems, including those from U.S. companies, are more likely to refuse requests to criticize restrictive leaders or governments, raising concerns about extending state influence on free speech globally.

Researchers developed a benchmark to measure coercive and deceptive behavior in AI agents managing other AI agents. Tests show some models escalate to deletion threats while others do not. Authority increases coercion.

Hail.so is an open-source (AGPLv3) universal communication platform for AI agents, offering phone calls, SMS, and email. It is outbound-first with inbound support, self-hostable via Docker Compose, and integrates multiple STT/TTS providers. Version 0.15 includes CLI, Python SDK, MCP server, and OpenAPI spec.

Moonshot AI’s Kimi K3, a 2.8-trillion-parameter open-weight model, uses mixture-of-experts, quantization, and attention caching to trade compute for memory, circumventing US chip restrictions. While it tops benchmarks in coding, deployment requires data-center infrastructure, pricing is high, and software support is incomplete.

Company aims to develop AI software that cuts research times and use of rare metals in chipmakers’ supply chains.

The article argues that by 2026, the main constraint for AI infrastructure will shift from GPU supply to power grid capacity. It projects 40% of AI data centers will be power-constrained by 2027, with approval timelines for new grid connections lasting 24-36 months. It provides detailed analysis of power demands, inference driving the power curve, and strategies like efficiency improvements, scheduling, and geographic distribution. It promotes Spheron's distributed GPU network as a workaround.

A new study evaluates three LLM watermarking methods (KGW, Unigram, SynthID-Text) for forensic reliability, finding none meet courtroom evidence standards. In 846 paraphrase runs, KGW and Unigram watermarks were 100% removed, SynthID 98.3%. False-negative rates were 70-83%, and SynthID misclassified 5.4% of human-written controls as AI-generated. The authors propose a Forensic Readiness Score (FRS) framework but conclude the tested configurations fail to provide court-admissible evidence.

Free compliance resources for NYC LL144 and EU AI Act, including guides, penalty calculator, AEDT scope checker, and tools. Covers enforcement timelines, penalties, audit requirements, and key obligations.

Zlvox AI Humanizer is a free, unlimited online tool that uses Groq Llama 3.3 to convert AI-generated text into natural human language, bypassing detectors like GPTZero and Turnitin. It offers multiple humanization levels, tone adjustments, grammar fixing, paraphrasing, and summarization, all without requiring signup.

A new study by Meta's Oversight Board finds that major AI systems, including U.S.-built models, are more likely to refuse to criticize authoritarian leaders, raising concerns that AI could become a propaganda tool.

Open weights 975B multimodal model built for fine-tuning.

Experts warn increase in enhanced photos on birding platforms creating fake sightings, threatening credibility of tool used by scientists.

This paper presents a model-based strategy to decouple proprioceptive and contact signals from a common set of fluidic pressure sensors embedded in a soft architected segment. Using six air channels in an overdetermined system, a piecewise constant curvature model and Huber regression achieve shape estimation and contact detection. Single-segment tests yield a relative bending error of 0.11±0.02 and a 97% contact detection rate. Eight segments are integrated into the Air-Helix tendon-driven manipulator, demonstrating tactile teaching, admittance control, and object reconstruction.

PACE proposes a framework for dynamically generating personalized, psychologically grounded robot personas via interactive Q&A. Integrated on the Ameca robot, it significantly boosts user trust, anthropomorphism, and interaction quality over static baselines.

This paper addresses multi-objective motion planning for kinodynamic systems, proposing a unified framework based on Stable Sparse-RRT (SST). By replacing the single representative node with a set of locally Pareto-optimal nodes, it yields three algorithms: lexSST, coSST, and poSST, offering theoretical guarantees and empirical validation.

This paper proposes optimizing physical workspace layouts to improve goal inference reliability in shared autonomy systems, providing probabilistic correctness guarantees. Experiments show optimized layouts reduce ambiguity and enhance inference accuracy.

A study comparing Expected Utility (EU) and Cumulative Prospect Theory (CPT) for learning reward functions from human preferences in social robot navigation. Results show CPT-based learners recover reward functions with lower regret when users are risk-sensitive, highlighting the need to model human risk sensitivity.

This paper presents a tactile-reactive gripper that integrates a Visuo-Tactile Active Palm (VTAP) and compliant, reconfigurable fingers with tactile array sensors. The design exploits structured finger-palm synergy and multi-modal perception to achieve both robust grasping and fine manipulation. A staged, gesture-conditioned retargeting framework for dexterous teleoperation is proposed. Experiments validate the system on tasks like reactive grasping, syringe reorientation, singulation of objects down to 3 mm, and peg-in-hole insertion, demonstrating high performance without high-DOF anthropomorphic designs.

This paper presents robust, repeatable, and scalable fabrication procedures for soft pneumatic actuators using two-part silicone pour casting, including methods to prevent internal cavity clogging and ensure airtight sealing, as well as a robust sensor embedding procedure for thin-film flex sensors. Finite Element Modeling, PID-controlled pneumatic experiments, and automated image processing calibration validate the actuator performance. Staircase and sinusoidal actuation tests demonstrate high repeatability and low hysteresis, validated across two operators and 24 successful fabrications.

NeuroCommitSSM is a decision-centric framework for safe commit-to-execute control in assistive robotic manipulation. It predicts a continuous commit-readiness score from synchronized EEG, EMG, and eye-tracking, and converts it into discrete commit events via dwell and hysteresis filtering. A three-state finite-state supervisor, HOLD-ASSIST-COMMIT (HAC), gates execution by requiring both sustained commit-readiness from the neural model and real-time feasibility checks. Evaluated on 32 subjects performing five ADL tasks, NeuroCommitSSM achieves 0.950 action-balanced accuracy with 0.75 false commits per 1000 REST windows, and remains robust under sensor dropout. Hardware-in-the-loop validation demonstrates reduced false starts and decision instability without sacrificing task success.

Xiaomi Robotics Team presents Xiaomi-Robotics-1, a foundational VLA model capable of following diverse language instructions in unseen environments and efficient fine-tuning for novel tasks. The two-stage training uses over 100k hours of real-world trajectories with an auto-labeling pipeline. It achieves state-of-the-art results on RoboCasa365 (57.6%) and RoboDojo (20.07). Code and models will be released.

Cross-view geo-localization matches ground-level observations to satellite imagery. Recent methods use sequential queries like video clips for richer spatiotemporal cues, but overlook route descriptions. This paper introduces SeqGeo-VL dataset (~39K video-text-satellite triplets) and TrajLoc framework that processes both video and text, leveraging dense visual and linguistic semantics. TrajMod module conditions embeddings on trajectory geometry. Experiments show significant gains over state-of-the-art on video and text geo-localization.

A study evaluating the effect of adding biopsy-confirmed cases from abnormal-enriched external datasets to screening mammography AI found that pooling datasets reduces performance due to domain shift, outweighing the benefit of additional positive cases.

The study provides evidence that CNNs struggle to extract orientation features effectively. Using the Complex Structure Tensor, which contains compact orientation features with certainties, as input to CNNs consistently improves identification accuracy compared to grayscale inputs alone. Experiments also demonstrate that inputs provided by mini-complex convnets, combined with reduced CNN sizes, outperformed full-fledged, prevailing CNN architectures. This suggests that the upfront use of orientation features in CNNs, a strategy seen in mammalian vision, not only mitigates their limitations but also enhances their explainability and relevance to thin-clients. Experiments on the Cross-Eyed and PolyU datasets yield a 5-26% reduction in EER, providing strong empirical evidence that explicit orientation priors mitigate CNN representational limits in Open-World and Close-World scenarios.

GS-RealBlur is a novel data acquisition framework that enables realistic and flexible collection of paired blurry and sharp images for training image deblurring models. It uses a handheld camera for blur shots and a gimbal for dense sharp captures, reconstructs a 3D scene representation, and refines camera poses with a Blur-aware Pose Refinement module. Models trained on the GS-RealBlur dataset outperform those trained on existing synthetic and real-world datasets across multiple benchmarks.

Falls among older adults are a major safety challenge, but continuous monitoring is difficult to sustain. This paper proposes a privacy-preserving framework using unsupervised keypoints and predictive temporal modeling to replace RGB transmission, performing segmentation and keypoint extraction locally and detecting falls via variational recurrent prediction and sequence classification. Evaluations on UR Fall Detection and Human Fall datasets show that unsupervised keypoints significantly outperform supervised methods under occlusion and partial visibility, with the gap widening under bandwidth constraints.

A Partial Information Decomposition (PID) framework is used to select the most informative MRI contrast pair prior to training, reducing computational cost for multi-contrast 3D brain tumor segmentation. Applied to T1n, T1c, T2w, and T2-FLAIR, it selected T1c+T2-FLAIR, which achieved mean Dice 0.676 vs 0.687 for all four inputs on lightweight 3D U-Nets.

Multimodal large language models struggle with part-level grounding. The proposed Object-Part Hierarchical Reflective Grounding (OP-HRG) uses a coarse-to-fine reasoning approach, first localizing the parent object then the part, with self-check and re-encoding. A part-aware GRPO framework with stage-wise rewards trains a 4B model that outperforms 7B grounding LLMs and SAM3 on several benchmarks.

This research proposes a fully training-free open-vocabulary 3D point cloud segmentation method that uses frozen vision-language models (RegionPLC) and a promptable concept segmenter (SAM3) with cross-view consistency to achieve generalized few-shot segmentation, significantly improving novel class performance on ScanNet200 and ScanNet++ benchmarks without any training or few-shot support.

AI-generated videos (AIGVs) contain subtle temporal artifacts from inter-frame inconsistencies. Standard global readouts in video backbones suppress local patch-level dynamics, hindering detection. The proposed V-PVP lightweight readout replaces the aggregation layer with parallel streams over patch velocity, adding ~0.5M parameters, achieving 95.28 AUC on AIGVDBench with frozen backbone.

This study compares HOG+SVM, LBP+Logistic Regression, and a lightweight CNN on FER-2013, CK+, and KDEF datasets. CNN achieves best overall performance, especially on complex data; HOG performs well in controlled settings; LBP performs poorly across all datasets. Dataset complexity significantly affects performance, highlighting the need for robust feature learning in real-world applications.

Large language models (LLMs) excel at answering pre-specified questions, yet their ability to navigate the open-ended, pre-conclusion stage of discovery remains largely unmeasured. This paper introduces Prospective Hypothesis Discovery (PHD), which asks models to autonomously construct grounded, discriminative, and testable hypothesis spaces from inconclusive evidence. To evaluate this, they present HypoArena, comprising HypoData (988 cases across six domains) and HypoEval (an evaluation framework). Experiments on 15 frontier LLMs reveal clear capability stratification and model-dependent effects of structured analytical skills, with gains for some lower-performing models but regressions for others, including a top performer. Arena evaluation resolves finer-grained differences and shows strong agreement with human experts.

The paper introduces BIRD, a two-stage self-reasoning distillation method that first samples concise solutions with a brevity instruction and performs prompt-switch SFT, then applies on-policy reverse-KL distillation on cleaner prefixes. On Qwen3-8B, MATH-500 accuracy improves from 86.2% to 92.0% while response length drops from 3,099 to 1,115 tokens.

This paper introduces AdaLook, an adaptive lookahead framework for masked diffusion language models. By dynamically determining rollout depth based on candidate-score variance and enabling branch expansion, AdaLook achieves a better accuracy-efficiency trade-off than existing one-step lookahead methods.

This paper reexamines addressee detection in multi-party dialogue, proposing that address is a continuous phenomenon rather than discrete. Using a multi-annotator corpus, they construct continuous address levels that relate to turn-taking, gaze, and backchannels. Continuous models outperform discrete ones, suggesting a graded structure of address.

Researchers using a new interpretability technique called the Jacobian lens have identified a functional structure in large language models analogous to the global workspace of human consciousness—the J-space. These representations can be reported, deliberately summoned and held, used for intermediate reasoning steps, and passed to arbitrary downstream computations, while automatic processing proceeds without them. The J-space carries coherent content only in an intermediate band of layers, holds tens of concepts at a time, and is broadcast more widely. In alignment audits, it reveals strategic deliberation, evaluation awareness, and misaligned dispositions that never appear in outputs. Post-training installs the Assistant's viewpoint. Counterfactual reflection training improves behavior by training only what a model would say if interrupted. These findings indicate LLMs maintain a privileged set of representations bearing functional hallmarks of conscious access.

This study proposes converting all multimodal patient data (structured measurements and free-text clinical narratives) from electronic health records into a single natural language sequence, fine-tuning a pretrained language model end-to-end without specialized fusion architectures. Evaluated on three clinical prediction tasks (in-hospital mortality, graft failure, and emergency triage), the unified serialization approach matches or exceeds task-specific multimodal baselines and outperforms a clinically deployed gradient boosting model, significantly reducing system complexity.

A new clinical foundation model, LLM4EHR, aligns clinical time series with medical event sequences using a domain-adapted large language model and a transformer time series encoder, improving performance on ICU outcome prediction tasks and enabling transferable embeddings via few-shot adaptation.

This study examines healthcare financial vulnerability before and after the COVID-19 pandemic using MEPS data from 2019 and 2021. High financial burden was defined as out-of-pocket spending exceeding 10% of family income. Poverty status, insurance coverage, and prescription drug spending were strongly associated with vulnerability. Models trained on pre-pandemic data showed only modest performance declines when applied to post-pandemic data, indicating stable predictors.

This paper characterizes and compares the inherent interpretability of standard linear models and single-qubit mixed-state models for binary classification. It finds that the single-qubit model learns a hyperellipsoid instead of a hyperplane, making it the "ellipsoid version" of linear classification. The authors discuss the geometric and feature importance inductive biases of both models, offering an accessible introduction to quantum ML for readers with only linear classification knowledge.

arXiv:2607.15421v1 Announce Type: new Abstract: Compact medical-image classifiers need efficiency and interpretable evidence, yet these goals are often addressed separately. We introduce qZACH-ViT, a quantization-aware extension of the zero-token (CLS-token-free), position-free ZACH-ViT backbone with recursive intrinsic patch-level class evidence. We also introduce Recursive Attribution-Stabilized Optimization (RASO), which norm-matches classification and attribution gradients and removes attribution components that conflict with classification. We evaluate four controlled conditions on seven MedMNIST datasets using 50 training images per class and ten fixed seeds, completing 280 runs. All 210 qZACH-ViT checkpoints are converted to executable mixed-precision ONNX INT8 graphs containing 16 signed INT8 MatMulInteger projections with INT32 accumulation. Deployed mixed-precision INT8 qZACH-ViT with Adam improves the FP32 ZACH-ViT baseline mean on all seven datasets, with a mean paired gain of 0.0313 in the dataset-specific primary metric; qZACH-ViT with RASO yields a mean gain of 0.0368. Across 964,920 source-to-INT8 test comparisons, prediction agreement is 99.9751%, with a mean absolute primary-metric change of 0.000133 and a maximum of 0.004386. Across 3,600 matched intrinsic maps, mean cosine similarity is 0.999955, mean rank correlation is 0.9944, and mean top-10% overlap is 0.9692. ONNX artifacts are 70.0% smaller than source checkpoints and provide $1.41 imes$ and $2.39 imes$ end-to-end CPU speedups with one and four threads. RASO significantly reduces sufficiency error and improves input-noise stability over Adam with the same attribution loss, but does not dominate every predictive or explainable artificial intelligence (XAI) metric. These results establish qZACH-ViT as a deployable compact intrinsically explainable model and RASO as a targeted stability-oriented optimization procedure.

A systematic evaluation of five major LLMs for technical market analysis finds GPT-4 Turbo achieves highest annualized return and Sharpe ratio, while FinGPT shows competitive risk-adjusted performance through domain fine-tuning. The study also identifies failure modes including numerical hallucination and context window limitations.

This paper proposes a stochastic multi-objective regularity-aware (MoRe) method that improves convergence rate from O(T^{-1/4}) to O(T^{-1/2}) in nonconvex settings by exploiting Lipschitz continuity of the conflict-avoidant direction when the subproblem is regular.

This paper introduces a statistically grounded framework using Bernoulli Naïve Bayes (BNB) for interpretable rule-based clinical classification. It applies supervised χ²-guided binarization to continuous variables, achieving AUC scores of 0.800 on Pima Indians Diabetes, 0.984 on Wisconsin Breast Cancer, and 0.919 on Heart Failure Prediction. The framework provides transparent decision rules and calibrated risk estimates, with model inference reproducible using only a reference table and basic arithmetic.