Skip to content
gentic.news — AI News Intelligence Platform
Connecting to the Living Graph…
gentic.news

AI Research

Breaking AI research news: latest papers from arXiv, NeurIPS, ICML, and top labs. Track transformer architecture advances, reasoning breakthroughs, and scientific discoveries in machine learning and artificial intelligence.

AI ResearchFunding & BusinessProducts & LaunchesBig TechStartupsOpen SourcePolicy & EthicsOpinion & Analysis
A bar chart comparing job task categories, with programming/math highlighted as a small slice at 7.6%, while…
AI Research
85

Stanford/CMU Study: AI Agent Benchmarks Focus on 7.6% of Jobs, Ignoring Management, Legal, and Interpersonal Work

Researchers analyzed 43 AI benchmarks against 72,000+ real job tasks and found they overwhelmingly test programming/math skills, which represent only 7.6% of actual economic work. Management, legal, and interpersonal tasks—which dominate the labor market—are almost entirely absent from evaluation.

x.com/Mar 16, 2026/3 min read
researchai agentsbenchmarks
Diagram comparing standard and improved agentic RAG system modules, highlighting contextualization and…
AI Research
99

New Research Improves Agentic RAG Efficiency with Contextualization and De-duplication Modules

Researchers propose test-time modifications to agentic RAG systems, adding contextualization and de-duplication modules. Their best variant achieves 5.6% higher accuracy and 10.5% fewer retrieval turns, making complex question-answering more efficient.

arxiv.org/Mar 16, 2026/3 min read/Multi-Source
natural language processingai efficiencyrag systems
A bar chart titled 'Accuracy of LLMs after multiple fact updates in long context' shows declining performance across…
AI Research
78

New Research Diagnoses LLMs' Struggle with Multiple Knowledge Updates in Context

A new arXiv paper reveals a persistent bias in LLMs when facts are updated multiple times within a long context. Models increasingly favor the earliest version, failing to track the latest state—a critical flaw for dynamic knowledge tasks.

arxiv.org/Mar 16, 2026/3 min read/Multi-Source
llmsknowledge managementai research
A researcher interacts with a dashboard showing an LLM agent's reasoning steps and retrieved sources, with branching…
AI Research
76

InterDeepResearch: A New Framework for Human-Agent Collaborative Information Seeking

Researchers propose InterDeepResearch, an interactive system that enables human collaboration with LLM-powered research agents. It addresses limitations of autonomous systems by improving observability, steerability, and context navigation for complex information tasks.

arxiv.org/Mar 16, 2026/3 min read
human-in-the-loopinformation retrievalmarket intelligence
A bar chart comparing LLM agent robustness scores, with Qwen3-30B-A3B leading at 79.6% invariant responses under…
AI Research
85

Semantic Invariance Study Finds Qwen3-30B-A3B Most Robust LLM Agent, Outperforming Larger Models

A new metamorphic testing framework reveals LLM reasoning agents are fragile to semantically equivalent input variations. The 30B parameter Qwen3 model achieved 79.6% invariant responses, outperforming models up to 405B parameters.

arxiv.org/Mar 16, 2026/3 min read
large-language-modelsresearchbenchmarks
A computer monitor displays a split-screen video analysis with bounding boxes tracking a runner across frames…
AI Research
84

SPARROW: A New Method for Precise Object Tracking in Video AI Models

Researchers introduce SPARROW, a technique that improves how AI models track and identify objects in videos with greater spatial precision and temporal consistency. This addresses critical limitations in current video understanding systems.

arxiv.org/Mar 16, 2026/3 min read
multimodal modelscomputer visionretail technology
A diagram illustrating a data pipeline with labeled stages for error-prone input, transformation, and robust…
AI Research
74

From Garbage to Gold: A Theoretical Framework for Robust Tabular ML in Enterprise Data

New research challenges the 'Garbage In, Garbage Out' paradigm, proving that high-dimensional, error-prone tabular data can yield robust predictions through proper data architecture. This has profound implications for enterprise AI deployment.

arxiv.org/Mar 16, 2026/3 min read
ai-theoryenterprise-airetail-tech
Diagram of ReasonGR framework showing multi-step semantic reasoning process for numerical queries in financial QA…
AI Research
80

ReasonGR: A Framework for Multi-Step Semantic Reasoning in Generative Retrieval

Researchers propose ReasonGR, a framework to enhance generative retrieval models' ability to handle complex, numerical queries requiring multi-step reasoning. Tested on financial QA, it improves accuracy for tasks like analyzing reports.

arxiv.org/Mar 16, 2026/3 min read/Multi-Source
information retrievaldata analyticsai research
A magnifying glass over a glowing digital fingerprint on a circuit board, symbolizing AI search analysis and data trust
AI Research
72

Algorithmic Trust and Compliance: A New Framework for Visibility in Generative AI Search

A new arXiv study introduces Generative Engine Optimization (GEO), a framework for optimizing content for AI search engines. It finds AI exhibits a strong bias towards authoritative, third-party sources, making compliance and trust signals critical for visibility in regulated sectors.

arxiv.org/Mar 16, 2026/3 min read
digital marketingresearchdata strategy
A diagram showing a pyramid-shaped architecture with multiple layers, each containing smaller blocks labeled…
AI Research
79

Expert Pyramid Tuning: A New Parameter-Efficient Fine-Tuning Architecture for Multi-Task LLMs

Researchers propose Expert Pyramid Tuning (EPT), a novel PEFT method that uses multi-scale feature pyramids to better handle tasks of varying complexity. It outperforms existing MoE-LoRA variants while using fewer parameters, offering more efficient multi-task LLM deployment.

arxiv.org/Mar 16, 2026/3 min read
efficiencyllmfine-tuning

CRYSTAL Benchmark Reveals Universal Step-Disorder in MLLM…

AI Research
95

CRYSTAL Benchmark Reveals Universal Step-Disorder in MLLMs: No Model Preserves >60% of Reasoning Steps in Correct Order

Researchers introduce CRYSTAL, a 6,372-instance benchmark evaluating multimodal reasoning through verifiable steps. It reveals systematic failures in 20 tested MLLMs, including universal cherry-picking and disordered reasoning chains.

arxiv.org/Mar 16, 2026/3 min read/Widely Reported
reasoningmultimodal-airesearch
Researchers analyzing brain signal generation methods on a monitor showing EEG waveform patterns and benchmark…
AI Research
84

Survey Benchmarks Four Approaches to Synthetic Brain Signal Generation for BCI Data Scarcity

A comprehensive survey categorizes and benchmarks four methodological approaches to generating synthetic brain signals for BCIs, addressing data scarcity and privacy constraints. The authors provide an open-source codebase for comparing knowledge-based, feature-based, model-based, and translation-based generative algorithms.

arxiv.org/Mar 16, 2026/3 min read
researchsurveymachine-learning
A flowchart diagram showing hierarchical LLM reasoning steps for multi-table data retrieval, with table icons and…
AI Research
92

FGTR: A New LLM Method for Fine-Grained Multi-Table Retrieval

Researchers propose FGTR, a hierarchical LLM reasoning method for retrieving precise data from multiple, large tables. It outperforms prior methods by 18-21% on standard benchmarks, moving beyond simple similarity search to a more analytical approach.

arxiv.org/Mar 16, 2026/3 min read/Multi-Source
llm applicationsdata & analyticsai research
A diagram of NanoVDR's distillation pipeline with teacher model outputs flowing into a compact text encoder, showing…
AI Research
82

NanoVDR: A 70M Parameter Text-Only Encoder for Efficient Visual Document Retrieval

New research introduces NanoVDR, a method to distill a 2B parameter vision-language retriever into a 69M text-only student model. It retains 95% of teacher quality while cutting query latency 50x and enabling CPU-only inference, crucial for scalable search over visual documents.

arxiv.org/Mar 16, 2026/3 min read/Multi-Source
information retrievalefficiencycomputer vision
A data scientist adjusts parameters on a laptop screen displaying colorful AI data flow charts, surrounded by…
AI Research
71

Goal-Driven Data Optimization: Training Multimodal AI with 95% Less Data

Researchers introduce GDO, a framework that optimizes multimodal instruction tuning by selecting high-utility training samples. It achieves faster convergence and higher accuracy using 5-7% of the data typically required. This addresses compute inefficiency in training vision-language models.

arxiv.org/Mar 16, 2026/3 min read
efficiencycomputer visionmachine learning
Terence Tao, a Fields Medalist, gestures while speaking at a conference, with a chalkboard of linear algebra…
AI Research
85

Terence Tao: LLM Math is Simple Undergraduate Linear Algebra, But Why They Work Remains a Mystery

Fields Medalist Terence Tao explains that the mathematics to build and run LLMs is straightforward linear algebra. The real puzzle is why they perform unpredictably across tasks, a gap in theory for 'meso-scale' natural data.

x.com/Mar 15, 2026/3 min read
theoryllmsresearch
A lab technician places a glass tissue slide under a microscope, with digital protein map overlays and survival…
AI Research
95

Microsoft's GigaTIME AI Predicts Protein Maps from $5 Tissue Slides, Revealing 1,234 New Survival Correlations

Microsoft released GigaTIME, an AI model that predicts expensive protein maps from cheap tissue slides. Trained on 40M cells from 14,256 patients, it discovered 1,234 new protein-survival connections.

x.com/Mar 15, 2026/3 min read
computer visionmicrosofthealthcare ai
Researchers at a whiteboard diagramming AI delegation workflow with trust models and task handoff arrows
AI Research
97

Google DeepMind Proposes 'Intelligent AI Delegation' Framework for Dynamic Task Handoffs with Verifiable Trust

Google DeepMind researchers propose a formal framework for delegating tasks to AI agents, treating delegation as a structured process with dynamic trust models, verifiable proofs, and failure management. The system is designed to prevent over- or under-delegation and enable AI-to-AI task handoffs with clear accountability.

x.com/Mar 15, 2026/3 min read
multi-agentresearchai-safety
A humanoid robot with mechanical joints runs on a dark Beijing street at midnight, its metal legs striding forward…
AI Research
87

Beijing Humanoid Robots Train for Half-Marathon in Midnight Test Runs

Humanoid robots are reportedly training for a half-marathon in Beijing, conducting late-night test runs. The event, scheduled for one month from the report, represents a significant public endurance test for bipedal robotics.

x.com/Mar 15, 2026/3 min read
chinaroboticsbenchmark
Andrej Karpathy stands at a podium delivering a presentation, with a bar chart behind him showing AI risk levels for…
AI Research
85

Andrej Karpathy Analysis: AI Poses High Risk to 57 Million US Jobs, ~40% of Workforce

Andrej Karpathy's analysis concludes AI puts 57 million US workers at high to very high risk of negative job impact. This ~40% figure contextualizes recent tech layoffs and discussions around universal high income.

x.com/Mar 15, 2026/3 min read
ai policylabor economicsanalysis
Researchers in a lab analyze a diagram showing interconnected AI agents, with arrows highlighting O(n²)…
AI Research
85

Researchers Apply Distributed Systems Theory to LLM Teams, Revealing O(n²) Communication Bottlenecks

A new paper applies decades-old distributed computing principles to LLM multi-agent systems, finding identical coordination problems: O(n²) communication bottlenecks, straggler delays, and consistency conflicts.

x.com/Mar 15, 2026/3 min read
llmsresearchdistributed systems
Data chart comparing worker productivity metrics before and after AI tool adoption, showing increased task…
AI Research
85

WSJ Report: AI Tools Increase Work Intensity for 164,000 Tracked Workers, Not Reduce Workloads

A Wall Street Journal analysis of data from 164,000 workers shows AI tools are making jobs more intense rather than reducing workloads. The finding challenges the common productivity narrative around workplace AI adoption.

x.com/Mar 15, 2026/3 min read
researchsocietybusiness
Two bipedal robots with articulated arms walking side by side in a lab, surrounded by equipment and cables
AI Research
85

Stanford's Mobile ALOHA Robots Now Walk Autonomously, Marking Key Mobility Advance

Stanford's Mobile ALOHA robots, previously requiring human guidance for movement, have gained autonomous walking capabilities. This represents a significant step toward general-purpose mobile manipulation.

x.com/Mar 15, 2026/3 min read
mobilityroboticsresearch
A laptop screen displays a data dashboard ranking job titles by AI exposure scores, with a bar chart showing wage…
AI Research
85

Andrej Karpathy's Deleted Tool: AI Exposure Scores for 342 Jobs, Finds $3.7T in High-Risk Wages

Andrej Karpathy briefly released a tool scoring 342 job types for AI exposure using an LLM, finding an average score of 5.3/10. The analysis identified $3.7 trillion in annual wages at high exposure (7+), with software developers at 9/10 and medical transcriptionists at 10/10.

x.com/Mar 15, 2026/3 min read
economicsresearchllm
A man in his 30s sits at a desk, typing on a laptop with a medical diagram of the spine on the screen, AI chatbot…
AI Research
85

Engineer Uses ChatGPT and Google to Self-Diagnose Rare Spinal Condition After 17-Month Medical Odyssey

A software engineer with no medical training used ChatGPT-4o and Google to correctly diagnose his own rare spinal CSF leak after 17 months of failed specialist consultations. The case highlights AI's emerging role as a diagnostic aid in complex medical scenarios.

x.com/Mar 15, 2026/3 min read
large language modelsmedical diagnosticshealthcare ai
Three 3D-printed robotic hands with visible tendons and joints, one with fingertip sensors, displayed on a workbench…
AI Research
97

ORCA Dexterity Open-Sources Three 3D-Printable Robotic Hands with Self-Dislocating Joints for ~$2,200

ORCA Dexterity released STL files for three tendon-driven anthropomorphic robotic hands featuring self-dislocating joints for reliability. The OrcaHand Touch variant includes high-resolution fingertip sensors with 83 taxels per fingertip at 1mm resolution.

x.com/Mar 15, 2026/3 min read
research-toolsopen-sourcehardware
A white humanoid robot on a green court holds a tennis racket in its right arm, posed mid-swing with one leg lifted…
AI Research
97

Tsinghua & Peking University Researchers Train Humanoid Robot to Play Tennis Using Scattered, Imperfect Human Motion Clips

A team from Tsinghua, Peking University, and other labs taught a humanoid robot to play tennis using short, imperfect human swing clips instead of perfect match data. The system uses a physics simulator to correct errors, lowering the barrier for teaching robots complex physical tasks.

x.com/Mar 15, 2026/3 min read
roboticsresearchimitation-learning
AI agent with a glowing brain icon and a stream of colored data flowing from past experiences into a skill library…
AI Research
89

XSkill Framework Enables AI Agents to Learn Continuously from Experience and Skills

Researchers have developed XSkill, a dual-stream continual learning framework that allows AI agents to improve over time by distilling reusable knowledge from past successes and failures. The approach combines experience-based tool selection with skill-based planning, significantly reducing errors and boosting performance across multiple benchmarks.

x.com/Mar 14, 2026/3 min read
continual learningmachine learningai research
An office worker stares at dual monitors displaying complex data dashboards and AI assistant chat windows…
AI Research
85

The AI Productivity Paradox: How Automation Tools Are Intensifying Workloads Instead of Easing Them

New research tracking 164,000 workers reveals AI tools are increasing work intensity rather than reducing it. Employees fill saved time with additional tasks, leading to longer hours and decreased focus time. Only 3% of users achieve the optimal balance of AI assistance.

x.com/Mar 14, 2026/3 min read
future of workproductivitytechnology impact
A tech founder at a computer screen with genetic data and a dog beside him, designing a personalized mRNA cancer…
AI Research
97

AI-Powered Breakthrough: Sydney Founder Creates Personalized mRNA Cancer Vaccine for Dog

A Sydney tech founder used ChatGPT and AlphaFold genetic data to design a personalized mRNA cancer vaccine for his dog Rosie after traditional treatments failed. Within weeks, a major tumor shrank by approximately 50%, demonstrating how AI could accelerate personalized cancer therapies.

x.com/Mar 14, 2026/3 min read
veterinary-innovationai-in-medicinepersonalized-therapeutics