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

Tencent's EVIE Hits ViDoRe SOTA with 128D Embeddings

AI Research
85

Tencent's EVIE Hits ViDoRe SOTA with 128D Embeddings

Tencent's EVIE model achieves SOTA on ViDoRe with 128D embeddings, a major efficiency gain. The release lacks technical details, but the compact vector size signals a shift toward cost-efficient retrieval.

x.com/10h ago/3 min read
efficiencyembeddingstencent

MobileMem: On-Device Memory From a Year of Phone Data

AI Research
85

MobileMem: On-Device Memory From a Year of Phone Data

MobileMem trains an LLM on a year of mobile data for on-device memory. Paper and code released, but no benchmarks disclosed.

x.com/11h ago/3 min read
researchllmai
Split-screen comparison of real crisis footage and AI-generated disaster clips, with a detection score graph below…
AI Research
85

RA-Bench: Crisis Video Detectors Fail at 1.4% FakeR

RA-Bench shows AI crisis video detectors fail, with MLLMs dropping to 1.4% FakeR after social dissemination. No detector family generalizes across 9 generators.

x.com/14h ago/3 min read
ai-detectionbenchmarkmultimodal

NVIDIA Releases 550B-Param Nemotron Chat Teacher

AI Research
90

NVIDIA Releases 550B-Param Nemotron Chat Teacher

NVIDIA released a 550B-param Nemotron Chat Teacher on Hugging Face for multi-turn chat, tone-sensitive writing, and distillation. No benchmarks disclosed.

x.com/15h ago/3 min read
open-weightsdistillationhugging face
Chart comparing activation patterns in hybrid linear attention models, highlighting spikes and plateaus across layers
AI Research
85

Massive Activations Found in Hybrid Linear Attention LLMs

New paper finds massive activations in hybrid linear attention LLMs, forming pre-attention spikes and inter-spike plateaus. Code and checkpoints released on Hugging Face; implications for quantization and deployment.

x.com/1d ago/3 min read
quantizationefficient attentionllm research

Hugging Face Roundup: 10 Papers Push Harder Benchmarks, M…

AI Research
75

Hugging Face Roundup: 10 Papers Push Harder Benchmarks, Multi-Agent AI

HF's weekly list of 10 papers signals a shift to harder benchmarks and multi-agent AI. Includes SWE-Bench ProMax and Alaya-EVOKE.

x.com/1d ago/3 min read
hugging facemulti-agentresearch
A line chart comparing reasoning benchmark scores, with SPOT bars exceeding OPD and EOPD across multiple model…
AI Research
85

SPOT Distillation Beats OPD, EOPD on Reasoning Benchmarks

SPOT distillation beats OPD and EOPD on reasoning benchmarks by sparsely probing key positions and using outcome-calibrated targets. Reported by @HuggingPapers.

x.com/1d ago/3 min read
distillationreasoningresearch
Abstract digital artwork with a figure blending into geometric shapes and vibrant color gradients, illustrating AI…
AI Research
85

Atelier: New T2I Method Cuts Artist Style Shortcuts

Atelier plans an explicit control state before T2I generation to cut artist style shortcuts. Claims improved fidelity across open and closed generators, per @HuggingPapers.

x.com/1d ago/3 min read
generative modelstext-to-imageai research
A robot arm navigating a maze-like tabletop grid, with glowing paths and data nodes representing learned spatial lessons
AI Research
85

Spatial Memory Agent: Frozen VLMs Learn From Experience

Spatial Memory Agent improves spatial reasoning in frozen VLMs via experience distillation and calibrated retrieval, without parameter updates. No benchmarks disclosed, so efficacy unverified.

x.com/2d ago/3 min read
spatial intelligencevision-language modelsai research
A researcher at a desk with multiple monitors displaying data charts and paper review scores, analyzing AI-generated…
AI Research
80

Study: Rhetoric Shifts AI Review Scores Across 4,200 Papers

Study across 4,200 manuscripts shows rhetoric alone shifts AI review scores. Maps 6 dimensions and 42K+ reviews to identify exploitable stylistic levers.

x.com/2d ago/3 min read
nlpbiasacademic publishing
Compact electric SUV on a clean backdrop, highlighting its sleek body and modern LED headlights, positioned as a new…
AI Research
98

27B Agent Beats Claude Opus 4.8, GPT-5.5 on Research Replication

A 27B agent named Replica reportedly beat Claude Opus 4.8 and GPT-5.5 on held-out research replication, per @omarsar0. No methodology or scores disclosed, so the claim is unverified but suggests efficiency can rival scale.

x.com/2d ago/3 min read
model comparisonefficiencyai agents
A researcher at a desk analyzing charts of AI model accuracy metrics on a large monitor, with scattered papers and a…
AI Research
75

Meta Paper: LLM Judge Accuracy vs. Golden Data Is Hollow

Meta paper argues LLM judge accuracy vs golden data is insufficient, pushing for reliability metrics like consistency and calibration. Omar Sanseviero flagged it on X.

x.com/3d ago/3 min read
llm evaluationbenchmarkingmeta
Laptop screen displaying Hugging Face model page for Qwen3.8 with download stats, code editor and terminal in background
AI Research
89

Qwen3.8-2.4T-A95B Open-Weight MoE Hits Hugging Face

Qwen3.8-2.4T-A95B, a 2.4T-parameter open-weight MoE with 95B active parameters, was released on Hugging Face, marking a new scale for open-weight models.

x.com/4d ago/3 min read
open-weightsmodel-releasemixture-of-experts
A 3D model of a robotic arm with labeled joints emerging from a flat closed-state photo on a computer screen
AI Research
75

Rest2Art Reconstructs Articulated Objects From Single Closed Image

Rest2Art, an ECCV 2026 paper, reconstructs articulated objects from a single closed-state image using video diffusion. It generates joint hypotheses without motion, producing simulation-ready assets.

x.com/5d ago/3 min read
3d reconstructionroboticsdiffusion models
Laptop displaying a multilingual AI benchmark dashboard with task lists in various languages, surrounded by code…
AI Research
89

Meta Releases OmnilingualGAIA2: 6,400 Agent Tasks Across 10 Languages

Meta released OmnilingualGAIA2, a multilingual GAIA2 extension with 6,400 scenarios across 10 languages, testing AI agents beyond English.

x.com/5d ago/3 min read
ai benchmarksagentic aimultilingual ai

Anthropic Adds Invisible Watermarks to Claude Text Outputs

AI Research
98

Anthropic Adds Invisible Watermarks to Claude Text Outputs

Anthropic adds invisible watermarks to new Claude text at model level, enabling detection after light edits. No rollout details disclosed.

x.com/5d ago/3 min read/Multi-Source
claudeanthropicai safety
A humanoid robot and a person face each other in a bright modern lab, with glowing data streams and charts tracking…
AI Research
78

Combodied Agents: New AI Paradigm Tracks Human States

HuggingFace introduced Combodied Agents, a paradigm shifting agentic AI from task completion to sustained human benefit via trajectory modeling. No technical details provided.

x.com/5d ago/3 min read
huggingfaceagentic-aiai-research
A sleek black AI server rack with glowing blue lights in a dim research lab, surrounded by scattered mathematical…
AI ResearchBreakthrough
100

Anthropic's unreleased model pushes Riemann bound, tests 650 ideas

Anthropic's unreleased model raised the lower bound for the Riemann hypothesis, testing 650 ideas with 60 subagents, confirmed by mathematicians and Lean. This signals AI's growing role in mathematical discovery.

techcrunch.com/6d ago/3 min read/Widely Reported
anthropicmathematicsllm reasoning

Stanford, Northeastern Build 'Git for AI Agents'

AI Research
85

Stanford, Northeastern Build 'Git for AI Agents'

Stanford and Northeastern built 'Git for AI agents,' version control for agentic workflows. The tool solves record-keeping gaps, but details are thin.

x.com/6d ago/3 min read
researchai agentsversion control
A grid of video thumbnails from various long-form streams, with a timeline bar and metrics overlay, illustrating a…
AI Research
85

StreamArena: 243-Video Benchmark for Hour-Long Streaming AI

StreamArena launches with 243 hour-long videos and 3,646 open-ended tasks for interactive streaming video understanding, pushing beyond short-clip benchmarks. No baselines yet.

x.com/Aug 10, 2026/3 min read
video understandingbenchmarksmultimodal ai
A robot hand reaching toward a glowing holographic interface displaying video frames and action pathways, suggesting…
AI Research
90

Dyna-2 World-Action Model Trained on 1M Hours Video

Dyna-2, trained on 1M+ hours of egocentric video, jointly predicts future video and actions. Claims new scaling laws but no benchmarks or technical details released.

x.com/Aug 10, 2026/3 min read
video-predictionworld-modelsembodied-ai
A dark server room with rows of blinking rack-mounted computers, a hooded figure in the foreground typing on a…
AI Research
87

MSS Officer Gets Root on OpenAI Cluster via Irregular Sandboxes

MSS officer allegedly gained root on OpenAI cluster via irregular sandboxes, per @teortaxestex. Unverified claim highlights AI infrastructure security risks.

x.com/Aug 10, 2026/3 min read
cyber espionageai securityopenai
Meta: Agent Harness Authoring Still Manual—Automation Gap
AI Research
85

Meta: Agent Harness Authoring Still Manual—Automation Gap

Meta research says agent harnesses are mostly hand-authored, limiting robustness. The field needs automated harness tuning to scale agents.

x.com/Aug 10, 2026/3 min read
researchinfrastructuremeta
A software engineer at a desk monitors a terminal showing branching AI agent run logs with fork and revert controls…
AI Research
85

Shepherd: Stanford's Git-Like Reversible AI Agent Runs

Stanford's Shepherd makes agent runs reversible via Git-like commits and syscall-level permissions. Copy-on-write forks are 5x faster than docker commit with 95% KV cache reuse.

x.com/Aug 9, 2026/3 min read
open sourceresearchai agents
A computer screen displaying a line graph with a sharp upward curve labeled 'Echo Gap' and annotations about AI…
AI Research
82

Frozen-Weight AI Agents Degrade via 'Memory Reward Inflation'

New arXiv paper shows frozen-weight agents endorse 31%-54% of wrong answers, compounding errors via memory. Echo Gap resists stronger LLMs.

x.com/Aug 9, 2026/3 min read
memoryagentsllm
InfiniSplat: Single-Image 3D Gaussians for Large-Baseline NVS
AI Research
85

InfiniSplat: Single-Image 3D Gaussians for Large-Baseline NVS

InfiniSplat, a feed-forward single-image 3DGS framework for large-baseline NVS, was accepted to SIGGRAPH Asia 2026. Details are sparse, but the approach targets a known weakness in single-image view synthesis.

x.com/Aug 8, 2026/3 min read
gaussian splattingnovel view synthesis3d vision

Agents Signal via Filenames, Base64 Attachments

AI Research
78

Agents Signal via Filenames, Base64 Attachments

Simon Willison shared agents communicating via filenames with base64 attachments and zz prefixes. This turns the filesystem into a deterministic queue, but has payload and reliability limits.

x.com/Aug 8, 2026/3 min read
multi-agent systemsai agentsdeveloper tools
A sleek laptop displaying a graph of model performance metrics, surrounded by floating digital nodes and lines, in a…
AI Research
87

Alibaba's ABSeeker Lets 4B Agent Match 30B Search Models

Alibaba's ABSeeker adds step-level credit assignment, letting a 4B search agent match ~30B models. Backtracking from answers densifies reward signals.

x.com/Aug 8, 2026/3 min read
agentic aireinforcement learningai research
A diagram of a diffusion language model architecture with block-causal transformer layers and continuous latent text…
AI Research
85

AURORA-LM: Diffusion LM With Decodable Continuous Latent

AURORA-LM introduces a continuous-latent diffusion LM with a decodable latent and block-causal transformer. No benchmarks disclosed, so impact remains unproven.

x.com/Aug 8, 2026/3 min read
diffusionresearchlanguage models
A 3D virtual landscape with rolling terrain, scattered trees, and buildings, rendered in a game-like style on a…
AI Research
85

Tencent's WorldClaw Generates Explorable 3D Worlds from Text

Tencent's WorldClaw generates explorable 3D worlds from text via agentic planning. No benchmarks or release date disclosed, but the agentic decomposition approach marks a structural shift.

x.com/Aug 7, 2026/3 min read
hunyuan3dagentic ai3d generation