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
Researchers in lab coats examining a glowing photonic chip under a microscope at Shanghai's new integrated photonic…
AI ResearchBreakthrough
95

China Launches Photonics Lab to Bypass US Chip Curbs on AI

China launched a photonics lab to bypass US chip curbs and develop energy-efficient AI computing using light instead of electrons.

scmp.com/Jun 12, 2026/3 min read/Widely Reported
ai hardwarephotonicssemiconductors
MiniMax M3 Exceeds Human Gold-Medal on Math Benchmarks via MaxProof
AI Research
100

MiniMax M3 Exceeds Human Gold-Medal on Math Benchmarks via MaxProof

MiniMax's M3 exceeded human gold-medal on math benchmarks via MaxProof, but no scores or details were disclosed.

x.com/Jun 12, 2026/3 min read/Multi-Source
reasoningmathematicsai
Bar chart with two groups labeled 'Rejected' and 'Accepted' comparing prediction scores, with a ROC curve inset…
AI Research
72

Clinical LLM Rejection Predictor Hits AUROC 0.719 in 4.5-Month Study

Clinical LLM rejection predictor achieves AUROC 0.719 in 4.5-month study using deployment-specific context to forecast user rejection before response generation.

arxiv.org/Jun 12, 2026/3 min read
llm evaluationarxivhealthcare

SVoT Boosts MLLM Spatial Reasoning by 65% via RL-Verified…

AI Research
88

SVoT Boosts MLLM Spatial Reasoning by 65% via RL-Verified Visual Chains

SVoT uses RL to verify MLLM spatial reasoning states, achieving up to 65% accuracy gains on OOD tests across five domains including Pacman and Gather.

arxiv.org/Jun 11, 2026/3 min read/Widely Reported
spatial-reasoningresearchreinforcement-learning

Lung-R1-14B Tops EMR Diagnosis with Knowledge Graph-Guide…

AI Research
92

Lung-R1-14B Tops EMR Diagnosis with Knowledge Graph-Guided RL

Lung-R1-14B scored 4.3583 on EMR diagnosis, beating 20 systems using a 59K-node knowledge graph and RL-constrained reasoning.

arxiv.org/Jun 11, 2026/3 min read/Widely Reported
researchaihealthcare
A sleek AI interface on a dark screen displays code analysis and exploit generation results, with a glowing network…
AI ResearchBreakthrough
84

Anthropic: Mythos Preview Builds Working Exploits in Hours, Not Weeks

Anthropic's Mythos Preview AI built 8 working exploits from Firefox and Windows kernel patches within hours. The first exploit was ready 18 days before the patched Firefox shipped.

the-decoder.com/Jun 10, 2026/3 min read
claudeanthropicai security
Two researchers point at a whiteboard covered with code and diagrams, illustrating AI agent orchestration with…
AI Research
100

Stanford, Meta 'Code as Agent Harness' Paper Rethinks AI Agent Design

Stanford and Meta's "Code as Agent Harness" paper proposes code-driven AI agent orchestration, potentially improving reliability over natural language prompts.

x.com/Jun 10, 2026/3 min read/Widely Reported
metaai agentsstanford
KV Cache Quantization Silently Breaks Safety Alignment, Paper Shows
AI Research
79

KV Cache Quantization Silently Breaks Safety Alignment, Paper Shows

KV cache quantization silently breaks LLM safety alignment, with Mistral-7B losing 15.2% refusals at 1.03x perplexity. PCR diagnostic recovers up to 97% alignment in 35 GPU-minutes.

arxiv.org/Jun 10, 2026/3 min read
alignmentllm inferenceai safety
Diagram of Visual-SDPO framework showing code-to-image pipeline with self-distillation feedback loop improving chart…
AI Research
68

Visual-SDPO: Self-Distillation Fixes Code-Generated Visual Defects by +10 Points

Visual-SDPO uses visual-feedback self-distillation to improve code-generated visual artifacts by >10 points on ChartMimic, Design2Code, and AeSlides, with no added inference cost.

arxiv.org/Jun 10, 2026/3 min read
code generationcomputer visionresearch
A 3D artist optimizes rendering settings in V-Ray software, adjusting parameters to speed up render times on a scene…
AI Research
92

Mirage: Microsoft's 10.57x faster video gen skips RGB render loop

Microsoft's Mirage stores 3D scenes as latent tokens, achieving 10.57x faster video generation and 55x less memory, with SOTA WorldScore consistency.

x.com/Jun 9, 2026/3 min read
3dresearchmicrosoft
A GPU chip with glowing circuits and a data flow diagram showing sparse attention patterns reducing KV cache load
AI Research
98

DeepSeek-V4 Hits 500K Context with 90% Less KV Cache via FlashMemory

DeepSeek-V4 achieves 500K context with 90% less KV cache via FlashMemory's lookahead sparse attention, keeping only 13.5% of cache in GPU memory without retraining.

x.com/Jun 9, 2026/3 min read/Multi-Source
efficiencyai modelsinference
A flowchart diagram mapping five AGI levels from responder to ecosystem, with arrows connecting stages of AI progress
AI Research
94

111-Page Survey Maps 5 AGI Levels: Responder to Ecosystem

111-page survey from US/China labs defines 5 AGI levels, argues epistemic exploration — not better answering — is key. Challenges scaling orthodoxy.

x.com/Jun 9, 2026/3 min read
explorationresearch paperagi
AI agent interface displaying Ebola sequence retrieval results with low counts, indicating missed data and altered…
AI Research
87

Anthropic: AI agents fail biology retrieval, miss 261 Ebola sequences

Anthropic research shows Claude Sonnet 4 returning 5–106 Ebola sequences instead of 266, shifting outbreak origin from 2014 to 1922. Repeatable retrieval tool fixes the variance.

x.com/Jun 8, 2026/3 min read
agent reliabilityretrieval-augmented generationbiology
Bar chart comparing AI models on the Artificial Analysis Intelligence Index, with MiniMax-M3 scoring 55 and leading…
AI Research
87

MiniMax-M3 Scores 55 on AI Index, Open-Source Lead Looms

MiniMax-M3 scored 55 on the Artificial Analysis Intelligence Index, set to become the leading open-source model once weights are released.

x.com/Jun 8, 2026/3 min read
open sourcebenchmarksai models
iPhone 17 Pro centered on a desk, displaying a multimodal AI interface, highlighting the sparse, advanced AFM Core…
AI Research
83

Apple AFM Core Advanced: Sparse, Multimodal, iPhone 17 Pro Only

Apple AFM Core Advanced is sparse, multimodal, and exclusive to iPhone 17 Pro, M3+ Mac, M4+ iPad, while AFM Core is dense for other devices.

x.com/Jun 8, 2026/3 min read/Multi-Source
hardware lock-inapplemultimodal
A line chart comparing small, medium, and large AI models shows the large model retaining rare skills longer during…
AI Research
88

Larger models learn rare skills by forgetting them less, new paper shows

New paper from Stanford, MIT, Harvard, and Anthropic shows larger models learn rare skills because they forget them less during training, tested on OLMo models from 4M to 4B parameters.

x.com/Jun 8, 2026/3 min read
anthropicscaling lawsmachine learning
WorldBench: Top MLLM Scores 64% on Visually Diverse Benchmark
AI Research
92

WorldBench: Top MLLM Scores 64% on Visually Diverse Benchmark

WorldBench, a new multimodal benchmark, tests 15 MLLMs on visually diverse images. Top model scores 64.0%, exposing fundamental gaps in visual understanding.

arxiv.org/Jun 8, 2026/3 min read/Widely Reported
computer visionbenchmarkmultimodal
A line graph titled 'Selective Attackers Cut Agent Safety by 28pp' shows a steep downward trend from 100% to 72%…
AI ResearchBreakthrough
100

Selective Attackers Cut Agent Safety by 28pp, Paper Finds

Strategic attack timing cuts agent AI safety by up to 28pp, showing current evaluations overestimate safety.

arxiv.org/Jun 8, 2026/3 min read/Widely Reported
research paperai safetyagentic ai
A macOS Activity Monitor window showing CPU, memory, and disk usage graphs, with multiple running processes listed below
AI Research
95

MacArena: 421-Task macOS Benchmark Reveals 26% CUA Ranking Inversion

MacArena benchmark of 421 macOS tasks reveals 26% performance gap for top models on native tasks, suggesting current CUAs overfit to Linux distributions.

arxiv.org/Jun 8, 2026/3 min read/Widely Reported
computer-use agentsapplereinforcement learning
Bar chart comparing weekly token growth of Chinese and US LLMs on OpenRouter, showing Chinese models surging past US…
AI Research
100

Chinese LLMs Surge on OpenRouter as U.S. AI Traffic Shifts

Chinese LLMs now drive most weekly token growth on OpenRouter, with American startups routing more traffic to them, per @rohanpaul_ai. The shift reflects utility over brand loyalty.

x.com/Jun 8, 2026/3 min read/Widely Reported
open sourceai infrastructurestartups
Diagram comparing standard MoE with 69.5 active experts to dMoE with 14.6, showing performance retention and memory…
AI Research
85

dMoE Cuts Active Experts from 69.5 to 14.6, Retains 99.11% Performance

dMoE reduces active experts from 69.5 to 14.6 in diffusion LLMs, retaining 99.11% performance while cutting memory 80% and speeding inference 1.66×.

x.com/Jun 7, 2026/3 min read
inferencemodel optimizationai research
Researchers analyzing a complex categorical framework diagram on a large screen, with mathematical symbols and…
AI Research
87

MIT Paper Formalizes Self-Revising AI Scientists That Can Change Their Own Language

MIT paper 2606.01444 formalizes self-revising AI scientists that can change their conceptual schema. Novelty is defined by what could not be expressed in the previous framework.

x.com/Jun 6, 2026/3 min read
agentic aiscientific discoveryai research

Google Titan: A New Architecture That Could Dethrone Tran…

AI Research
87

Google Titan: A New Architecture That Could Dethrone Transformers

Google's Titan architecture claims to surpass Transformers on long-context tasks via neural long-term memory, achieving 1.2x-2.5x speedups on benchmarks.

x.com/Jun 6, 2026/3 min read
neural networkslong-contextgoogle
Claude Opus 4.7 Matches Dedicated NMR Software on Chemistry Tasks
AI Research
94

Claude Opus 4.7 Matches Dedicated NMR Software on Chemistry Tasks

Claude Opus 4.7 matches NMR software on chemistry tasks per Anthropic blog, but methodology and benchmarks undisclosed.

x.com/Jun 5, 2026/3 min read
anthropicscientific aiai research
SMAC-Talk: StarCraft Benchmark Tests LLM Agents Against Deceptive Allies
AI Research
70

SMAC-Talk: StarCraft Benchmark Tests LLM Agents Against Deceptive Allies

SMAC-Talk extends StarCraft Multi-Agent Challenge with natural language communication, testing LLM agents against deceptive allies. Qwen3.5 models benchmarked; no model exceeds 72% win rate.

arxiv.org/Jun 5, 2026/3 min read/Multi-Source
ai safetyai agentsbenchmarks
Two AI model performance comparison charts side by side, one labeled Nemotron 3 Ultra and the other GPT-5.5, with…
AI Research
85

Nemotron 3 Ultra matches GPT-5.5 on physics test at 10X lower cost

Nemotron 3 Ultra matched GPT-5.5 on a physics test at 10X lower cost ($0.051 vs $0.57), highlighting MoE efficiency.

x.com/Jun 5, 2026/3 min read
moeai modelsinference
Anthropic engineer reviewing code on dual monitors with Claude AI interface displaying merged code contributions and…
AI Research
99

Anthropic: Claude Authors 80%+ of Code, Task Length Doubling Every 4 Months

Anthropic reports Claude authors 80%+ of code; task-length capability doubles every 4 months. Mythos Preview works 16+ hours autonomously.

x.com/Jun 4, 2026/3 min read
ai capabilitiescode generationclaude
A computer screen displays a complex web page with hidden code snippets highlighted in red, while a robotic hand…
AI Research
100

DeepMind paper: hidden web content hijacks agents 86% of the time

DeepMind catalogues 6 attack types where hidden web content hijacks AI agents up to 86% of the time, reframing safety from model alignment to environment trust.

x.com/Jun 4, 2026/3 min read/Multi-Source
research paperai securityautonomous agents
Two AI agent icons face a regulatory checklist document, with a 48.3% coverage score displayed above, representing…
AI Research
88

Ontology-Grounded AI Agent Testing Hits 48.3% Regulatory Coverage vs.

Ontology-grounded AI agent testing achieves 48.3% regulatory coverage vs. 33.1% baseline in 1800-scenario pilot. Coverage advantage over RAG not robust after Bonferroni correction.

arxiv.org/Jun 4, 2026/3 min read/Widely Reported
ai agentsregulatory complianceenterprise ai

Law Profs Prefer AI Answers 75% of Time in Stanford Study

AI Research
85

Law Profs Prefer AI Answers 75% of Time in Stanford Study

Stanford researchers found law professors preferred AI answers 75% of time in blind legal analysis test, per @rohanpaul_ai.

x.com/Jun 3, 2026/3 min read
legalresearchai