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
Screenshot of a web browser interface showing MolmoWeb AI agent navigating a YouTube page, with highlighted action…
AI ResearchBreakthrough
100

AI2's MolmoWeb: Open 8B-Parameter Web Agent Navigates Using Screenshots, Challenges Proprietary Systems

The Allen Institute for AI released MolmoWeb, a fully open web agent that operates websites using only screenshots. The 8B-parameter model outperforms other open models and approaches proprietary performance, with all training data and weights publicly released.

the-decoder.com/Mar 25, 2026/3 min read/Multi-Source
open sourcecomputer visionai agents
Bar chart comparing Claude 3.5 Sonnet and Claude 3 Opus usage growth over two months, with Sonnet showing a 50% increase
AI Research
97

Anthropic's Economic Index: Claude 3.5 Sonnet Usage Grows 50% After 2 Months, Outpacing Claude 3 Opus

Anthropic's first Economic Index shows users who adopt Claude 3.5 Sonnet increase their usage by 50% after two months, while Claude 3 Opus usage grows 20%. The data suggests Sonnet's efficiency drives deeper integration into workflows.

x.com/Mar 25, 2026/3 min read/Multi-Source
anthropicbusiness intelligencellm
A digital illustration of interconnected AI nodes forming a network, with a central glowing brain symbolizing…
AI Research
95

Meta's Hyperagents Enable Self-Referential AI Improvement, Achieving 0.710 Accuracy on Paper Review

Meta researchers introduce Hyperagents, where the self-improvement mechanism itself can be edited. The system autonomously discovered innovations like persistent memory, improving from 0.0 to 0.710 test accuracy on paper review tasks.

x.com/Mar 25, 2026/3 min read
ai-agentsresearchmeta-learning
A diagram comparing OCR model architectures, showing a diffusion model with parallel block-wise decoding versus…
AI Research
85

MinerU-Diffusion: A 2.5B Parameter Diffusion Model for OCR Achieves 3.2x Speedup Over Autoregressive Methods

Researchers introduced MinerU-Diffusion, a 2.5B parameter diffusion model for OCR that replaces autoregressive decoding with parallel block-wise diffusion. It achieves up to 3.2x faster inference while improving robustness on complex documents with tables and formulas.

x.com/Mar 25, 2026/3 min read
diffusion-modelsdocument-understandingperformance
Two Google DeepMind researchers examining a large language model interface on a monitor, with conversation bubbles…
AI Research
85

Google DeepMind's 'Learning Through Conversation' Paper Shows LLMs Can Improve with Real-Time Feedback

Google DeepMind researchers have published a paper demonstrating that large language models can be trained to learn and improve their responses during a conversation by incorporating user feedback, moving beyond static pre-training.

x.com/Mar 25, 2026/3 min read
researchgooglelarge language models
Three stylized robot heads with glowing blue eyes positioned around a central glowing geometric symbol, representing…
AI Research
85

LLM Multi-Agent Framework 'Shared Workspace' Proposed to Improve Complex Reasoning via Task Decomposition

A new research paper proposes a multi-agent framework where LLMs split complex reasoning tasks across specialized agents that collaborate via a shared workspace. This approach aims to overcome single-model limitations in planning and tool use.

x.com/Mar 25, 2026/3 min read
architecturereasoningagents
A schematic diagram of a web agent system, showing components like perception, reasoning, and action modules…
AI Research
95

Ego2Web Benchmark Bridges Egocentric Video and Web Agents, Exposing Major Performance Gaps

Researchers introduce Ego2Web, the first benchmark requiring AI agents to understand real-world first-person video and execute related web tasks. Their novel Ego2WebJudge evaluation method achieves 84% human agreement, while state-of-the-art agents perform poorly across all task categories.

arxiv.org/Mar 25, 2026/3 min read/Widely Reported
ai-agentsresearchbenchmarks

VHS: Latent Verifier Cuts Diffusion Model Verification Co…

AI Research
95

VHS: Latent Verifier Cuts Diffusion Model Verification Cost by 63.3%, Boosts GenEval by 2.7%

Researchers propose Verifier on Hidden States (VHS), a verifier operating directly on DiT generator features, eliminating costly pixel-space decoding. It reduces joint generation-and-verification time by 63.3% and improves GenEval performance by 2.7% versus MLLM verifiers.

arxiv.org/Mar 25, 2026/3 min read/Widely Reported
efficient aicomputer visionresearch
A surgeon's gloved hands use instruments during a laparoscopic procedure, with a monitor displaying the internal…
AI Research
75

Halsted VLM: A 650,000-Video Surgical Atlas and Platform for Temporal Procedure Mapping

Researchers introduce Halsted, a vision-language model trained on over 650,000 annotated surgical videos across eight specialties. It surpasses prior SOTA in mapping surgical activity and is deployed via a web platform for direct surgeon use.

arxiv.org/Mar 25, 2026/3 min read
healthcare-airesearchcomputer-vision
A neural network diagram showing a Vision Transformer processing sequential low-resolution image glimpses, with…
AI Research
91

CanViT: First Active-Vision Foundation Model Hits 45.9% mIoU on ADE20K with Sequential Glimpses

Researchers introduce CanViT, the first task- and policy-agnostic Active-Vision Foundation Model (AVFM). It achieves 38.5% mIoU on ADE20K segmentation with a single low-resolution glimpse, outperforming prior active models while using 19.5x fewer FLOPs.

arxiv.org/Mar 25, 2026/3 min read/Multi-Source
active-perceptiontransformerresearch
Researchers Yann LeCun and team stand before a screen displaying LeWorldModel's scene dynamics, a 15M-parameter AI…
AI Research
87

LeWorldModel: Yann LeCun's Team Achieves Stable World Model Training with 15M Parameters, No Training Tricks

Researchers including Yann LeCun introduce LeWorldModel, a 15M-parameter world model that learns scene dynamics from raw pixels without complex training stabilization tricks. It trains in hours on one GPU and plans 48x faster than foundation-model-based alternatives.

x.com/Mar 25, 2026/3 min read
roboticsresearchmeta
Google's TurboQuant Cuts LLM KV Cache Memory by 6x, Enables 3-Bit Storage Without Accuracy Loss
AI Research
95

Google's TurboQuant Cuts LLM KV Cache Memory by 6x, Enables 3-Bit Storage Without Accuracy Loss

Google released TurboQuant, a novel two-stage quantization algorithm that compresses the KV cache in long-context LLMs. It reduces memory by 6x, achieves 3-bit storage with no accuracy drop, and speeds up attention scoring by up to 8x on H100 GPUs.

x.com/Mar 25, 2026/3 min read
researchoptimizationinference
A researcher reviewing a complex diagram of AI research pathways on a large monitor, with highlighted trajectory…
AI Research
85

OpenResearcher Paper Released: Method for Synthesizing Long-Horizon Research Trajectories for AI

The OpenResearcher paper has been released, exploring methods to synthesize long-horizon research trajectories for deep learning. This work aims to provide structured guidance for navigating complex, multi-step AI research problems.

x.com/Mar 24, 2026/3 min read
researchmeta-learningai tools
Line graph comparing user task types over time, with iteration curve rising above autonomy curve, indicating shift…
AI Research
85

Anthropic Economic Index: Claude Users Shift from Autonomy to Iteration, Attempt Higher-Value Tasks

Anthropic's latest Economic Index data shows experienced Claude users increasingly prefer iterative collaboration over full autonomy, while attempting higher-value tasks with greater success rates.

x.com/Mar 24, 2026/3 min read
claudeanthropicai behavior
Engineers at a computer monitor displaying a multi-agent system orchestrating Claude instances, with code and UI…
AI Research
85

Anthropic Deploys Multi-Agent Harness to Scale Claude's Frontend Design & Autonomous Software Engineering

Anthropic engineers detail a multi-agent system that orchestrates multiple Claude instances to tackle complex, long-running software tasks like frontend design. The approach aims to overcome single-model context and reasoning limits.

x.com/Mar 24, 2026/3 min read
software developmentindustry trendsanthropic
NVIDIA CEO Jensen Huang on stage presenting a slide about synthetic data and compute bottlenecks in AI training
AI Research
85

Jensen Huang Predicts AI Training Shift to Synthetic Data, Compute as New Bottleneck

NVIDIA CEO Jensen Huang states AI training is moving from real-world to synthetic data, with compute power becoming the primary constraint as AI-generated data quality improves.

x.com/Mar 24, 2026/3 min read
industry trendssynthetic dataai hardware
A robotic arm with a two-fingered gripper reaching toward a cluttered table of household objects like mugs and…
AI Research
97

Meta's V-JEPA 2.1 Achieves +20% Robotic Grasp Success with Dense Feature Learning from 1M+ Hours of Video

Meta researchers released V-JEPA 2.1, a video self-supervised learning model that learns dense spatial-temporal features from over 1 million hours of video. The approach improves robotic grasp success by ~20% over previous methods by forcing the model to understand precise object positions and movements.

x.com/Mar 24, 2026/3 min read
self-supervised-learningroboticsresearch
Wharton researcher points at a graph on a screen showing declining review accuracy, while a colleague frowns at a…
AI Research
85

Wharton Study Finds 'AI Writes, Humans Review' Model Failing in Real Business Contexts

New Wharton research reveals the 'AI writes, humans review' workflow is breaking down in practice, with human reviewers struggling to effectively evaluate AI-generated content. The study suggests current review processes may be insufficient for quality control.

x.com/Mar 24, 2026/3 min read
human-ai collaborationbusiness aiworkflow

LLMs Can Now De-Anonymize Users from Public Data Trails, …

AI Research
85

LLMs Can Now De-Anonymize Users from Public Data Trails, Research Shows

Large language models can now identify individuals from their public online activity, even when using pseudonyms. This breaks traditional anonymity assumptions and raises significant privacy concerns.

x.com/Mar 24, 2026/3 min read
privacyai ethicsllms
Kimi 2.5's 1T Parameter MoE Model Runs on 96GB Mac Hardware via SSD Streaming
AI Research
85

Kimi 2.5's 1T Parameter MoE Model Runs on 96GB Mac Hardware via SSD Streaming

Developers have demonstrated that Kimi 2.5's 1 trillion parameter Mixture-of-Experts model can run on Mac hardware with just 96GB RAM by streaming expert weights from SSD, with only 32B parameters active per token.

x.com/Mar 24, 2026/3 min read
inference-optimizationmodel-deploymenthardware
A diagram illustrates the Mix-and-Match Pruning framework with Swin-Tiny layers, showing multiple pruning masks in…
AI Research
81

Mix-and-Match Pruning Framework Reduces Swin-Tiny Accuracy Degradation by 40% vs. Single-Criterion Methods

Researchers introduce Mix-and-Match Pruning, a globally guided, layer-wise sparsification framework that generates diverse pruning configurations by coordinating sensitivity scores and architectural rules. It reduces accuracy degradation on Swin-Tiny by 40% relative to standard pruning, offering Pareto-optimal trade-offs without repeated runs.

arxiv.org/Mar 24, 2026/3 min read
researchcomputer-visionmodel-optimization
A diagram showing six network impairment icons—latency, jitter, packet loss, bandwidth, reordering, and…
AI Research
95

AgentComm-Bench Exposes Catastrophic Failure Modes in Cooperative Embodied AI Under Real-World Network Conditions

Researchers introduce AgentComm-Bench, a benchmark that stress-tests multi-agent embodied AI systems under six real-world network impairments. It reveals performance drops of over 96% in navigation and 85% in perception F1, highlighting a critical gap between lab evaluations and deployable systems.

arxiv.org/Mar 24, 2026/3 min read/Widely Reported
roboticsai safetyresearch
A computer screen displays a complex AI framework diagram with nodes and connecting lines, representing the…
AI Research
91

AgenticGEO: Self-Evolving AI Framework for Generative Search Engine Optimization Outperforms 14 Baselines

Researchers propose AgenticGEO, an AI framework that evolves content strategies to maximize inclusion in generative search engine outputs. It uses MAP-Elites and a Co-Evolving Critic to reduce costly API calls, achieving state-of-the-art performance across 3 datasets.

arxiv.org/Mar 24, 2026/3 min read/Multi-Source
search enginesmachine learningoptimization
Researchers present DST, a plug-and-play predictor for Tree of Thought reasoning, with a diagram showing reduced…
AI Research
83

DST: Domain-Specialized Tree of Thought Cuts Computational Overhead by 26-75% with Plug-and-Play Predictors

Researchers introduce DST, a plug-and-play predictor that guides Tree of Thought reasoning with lightweight supervised heuristics. The method matches or exceeds standard ToT accuracy while reducing computational costs by 26-75% across mathematical and logical reasoning benchmarks.

arxiv.org/Mar 24, 2026/3 min read/Multi-Source
large-language-modelsreasoningefficiency
A diagram showing interconnected nodes representing specialized text-to-image models, with highlighted subgraphs…
AI Research
95

DiffGraph: An Agent-Driven Graph Framework for Automated Merging of Online Text-to-Image Expert Models

Researchers propose DiffGraph, a framework that automatically organizes and merges specialized online text-to-image models into a scalable graph. It dynamically activates subgraphs based on user prompts to combine expert capabilities without manual intervention.

arxiv.org/Mar 24, 2026/3 min read/Widely Reported
computer visionresearchgenerative ai
A researcher points at a whiteboard diagram showing 7 operators moving data between zones labeled 'black fog' and…
AI Research
80

Context Cartography: Formal Framework Proposes 7 Operators to Govern LLM Context, Moving Beyond 'More Tokens'

Researchers propose 'Context Cartography,' a formal framework for managing LLM context as a structured space, defining 7 operators to move information between zones like 'black fog' and 'visible field.' It argues that simply expanding context windows is insufficient due to transformer attention limitations.

arxiv.org/Mar 24, 2026/3 min read
ai-engineeringtransformer-architecturellms
A diagram comparing standard attention memory allocation with PagedAttention's block-based KV cache management…
AI Research
95

arXiv Survey Maps KV Cache Optimization Landscape: 5 Strategies for Million-Token LLM Inference

A comprehensive arXiv review categorizes five principal KV cache optimization techniques—eviction, compression, hybrid memory, novel attention, and combinations—to address the linear memory scaling bottleneck in long-context LLM inference. The analysis finds no single dominant solution, with optimal strategy depending on context length, hardware, and workload.

arxiv.org/Mar 24, 2026/3 min read/Widely Reported
llm-optimizationresearchsurvey
A diagram of an LLM with attention flow highlighting internal parameters, illustrating how it accesses its own…
AI Research
86

LLMs Show 'Privileged Access' to Own Policies in Introspect-Bench, Explaining Self-Knowledge via Attention Diffusion

Researchers formalize LLM introspection as computation over model parameters, showing frontier models outperform peers at predicting their own behavior. The study provides causal evidence for how introspection emerges via attention diffusion without explicit training.

arxiv.org/Mar 24, 2026/3 min read
emergencellmsresearch
Researchers pose beside a server rack, pointing at a screen displaying the Seed1.8 model architecture diagram, in a…
AI Research
99

Seed1.8 Model Card Released: A 1.8B Parameter Foundation Model for Generalized Real-World AI Agents

Researchers have introduced Seed1.8, a 1.8 billion parameter foundation model designed for generalized real-world agency. It maintains strong LLM and vision-language capabilities while adding unified interfaces for search, code execution, and GUI interaction.

arxiv.org/Mar 24, 2026/3 min read/Widely Reported
foundation modelsresearchai agents
A person in a dimly lit room gestures toward a glowing, abstract network of overlapping geometric shapes on a large…
AI Research
99

λ-RLM: 8B Parameter Model Using Typed λ-Calculus Beats 405B Performance on Long-Context Tasks

Researchers developed λ-RLM, an 8B parameter model that outperforms 405B models on long-context tasks by replacing recursive code with typed λ-calculus combinators. This approach guarantees termination and reduces latency by up to 4.1x.

x.com/Mar 24, 2026/3 min read
reasoningefficiencylanguage-models