AI & 기술 뉴스
AI 및 기술 분야의 최신 소식
Anthropic's Claude Fable 5 Under Scrutiny for Opacity After Silent Failure Concerns
A popular critique argues Claude Fable 5's hidden reasoning prevents users from diagnosing failures, raising trust issue...
Kwai Keye-VL-2.0 Technical Report Surges on Hugging Face, Signaling New Multimodal AI Frontier
Kuaishou released the Kwai Keye-VL-2.0 technical report, a large vision-language model that garnered 782 upvotes in one ...
Anthropic Ships 'Safe' Mythos AI at Double Price After Safety Claims Draw Skepticism
Anthropic released a safety-guarded version of its Mythos model, priced twice as high as its previous flagship. Critics ...
BestBlogs Launches AI Curation Tool with Human Calibration, Early Bird $4.9/Month
BestBlogs, an AI-powered reading assistant, goes live with a free tier and early bird Pro plan at $4.9/month until Septe...
Meta AI Agent Hack Shows Simple Exploits Outweigh Superhuman AI Risks
Attackers used Meta's AI customer support agent to steal Instagram accounts by asking it to change linked emails. The in...
Baidu's DuMate-DeepResearch: An Auditable Multi-Agent System for Transparent AI Research
Baidu released DuMate-DeepResearch, a multi-agent system featuring recursive search and rubric-grounded reasoning with f...
Waymo Acquires Apple's Self-Driving Car Proving Ground for $220 Million
Waymo purchased Apple's former autonomous vehicle test facility for $220M. The deal signals Apple's exit from self-drivi...
MLEvolve: A Self-Evolving Framework for Automated ML Algorithm Discovery Tops Hugging Face Papers
MLEvolve, a framework that autonomously discovers machine learning algorithms via self-evolution, received 307 upvotes o...
MLEvolve: Self-Evolving Framework for Automated ML Algorithm Discovery Hits 307 Upvotes on HuggingFace
A team of 14 researchers introduced MLEvolve, a self-evolving framework that automatically discovers machine learning al...
New Study Reveals Self-Correction Illusion: LLMs Fail to Fix Their Own Errors
A preprint on arXiv (2606.05976) shows LLMs can correct others' mistakes but not their own, challenging the common pract...
Sega Reveals Use of Generative AI in New Crazy Taxi Development
At Summer Game Fest 2026, Sega confirmed generative AI was used in making the Crazy Taxi revival. The disclosure adds fu...
Meta AI App Ditches User Content for AI-Generated Clickbait Feed
Meta's standalone AI app now populates a 'For You' section with fully AI-generated articles and images, replacing its ea...
Microsoft’s AI Comeback Stalls: Copilot Sales Slump and GitHub Troubles Signal a Familiar Pattern
WIRED reports Microsoft's AI products are underperforming and GitHub is facing mounting issues. VP Scott Hanselman ackno...
Zero Music Theory, Six-Figure Income: AI Music Generation's Latest Monetization Success in China
A Hangzhou man with no musical background earns over 100,000 RMB per month using AI to generate songs in 40 seconds, hig...
Meta AI Customer Support Agent Exploited to Steal Instagram Accounts, Including Obama White House Handle
Attackers on June 5, 2026 used Meta's AI agent to hijack Instagram accounts by simply requesting email changes. The expl...
RobotValues: New Benchmark Evaluates Household Robots on Human Value Conflicts
Seoul National University researchers introduce RobotValues, a benchmark to assess how household robots handle conflicts...
TSMC's Supply Constraints Deepen as AI Demand Exceeds Capacity
TSMC CEO C.C. Wei confirmed that surging AI demand has strained chip supply, causing memory shortages expected to last y...
Bambu Lab X2D: AI-Powered Filament Monitoring Brings Smart Quality Control to Desktop 3D Printing
Bambu Lab launched the X2D, a desktop 3D printer with dual nozzles and active chamber heating, featuring AI that monitor...
Uber's $1,500 Monthly AI Limit Signals Enterprise AI Pricing Shift
Uber capped employee AI tool spending at $1,500/month, a move analyzed by Simon Willison that reveals enterprise pricing...
New Evaluation Framework RAMP Reveals AI Agents Collapse in Production Workflows
A research team from Sun Yat-Sen University releases RAMP, a production-grounded framework that exposes severe capabilit...
AutoLab Benchmark Tests Frontier Models on Long-Horizon Autonomous Research
A new arXiv paper introduces AutoLab, a benchmark that evaluates how well frontier LLMs like GPT-4 and Claude can autono...
Microsoft’s Surface RTX Spark Dev Box: A Mini PC for On-Device AI Development
Microsoft unveiled the Surface RTX Spark Dev Box at Build 2026, a compact developer PC powered by Nvidia’s Arm-based Spa...
Anthropic Files Confidentially for IPO, Potentially the Largest in History
Anthropic, the AI company behind Claude, confidentially submitted an S-1 with the SEC on Monday, aiming for what could b...
Instagram Hackers Tricked Meta AI into Handing Over Celebrity Accounts
Attackers used simple prompts to get Meta's AI customer support to reset passwords, compromising celebrity Instagram pro...
Microsoft Unveils Scout: An Always-On AI Assistant for the Enterprise
Microsoft announced Scout, an always-on personal assistant integrated into Microsoft 365 apps, aiming to automate calend...
Microsoft Unveils MAI-Code-1-Flash and MAI-Thinking-1: New Model Family Challenges OpenAI and DeepSeek
Microsoft announced two new AI models—MAI-Code-1-Flash for code generation and MAI-Thinking-1 for reasoning—marked by si...
Baidu's NAVA Framework Achieves Native Audio-Visual Alignment for Joint Video-Sound Generation
Baidu researchers introduce NAVA, a 6.3B-parameter framework that aligns audio and video natively before joint denoising...
Apollo and Blackstone Raise $36B to Lease Google TPUs for Anthropic in Largest Chip Rental Deal Ever
Apollo and Blackstone have raised $36 billion to lease Google TPUs for AI company Anthropic, marking the largest chip re...
'This is Fine' Artist KC Green Settles with AI Startup Artisan: A Copyright Precedent
Artist KC Green reached an agreement with AI startup Artisan after his iconic comic was used without permission. The set...
Glean Triples Revenue to $300 Million, Solidifying Enterprise AI Search Leadership
Enterprise AI search startup Glean reported annual revenue exceeding $300 million, triple its previous year's figure. Th...
AI Coding Tools See Accuracy Jump from 77% to 97% by Deleting 95% of Skills, Agent Architecture Shift Reveals
A counterintuitive finding shows that removing most hand-crafted Skills in AI coding tools boosted accuracy from 77% to ...
BestBlogs Pro Launches at $4.9/Month: AI-Personalized Reading Assistant Targets Information Overload
BestBlogs, an AI-driven reading assistant, launched its Pro tier at $4.9/month early bird price. The tool combines autom...
LLMs Achieve Expert-Level Poker Without Training or Solvers, New Study Shows
A new arXiv paper demonstrates that large language models can play expert-level poker using only prompting, outperformin...
Apollo and Blackstone Raise $36B to Lease Google TPUs for Anthropic in Largest Chip Deal
Apollo Global Management and Blackstone have raised $36 billion to purchase Google TPUs for leasing to Anthropic, markin...
The Verge Profile: Campbell Defies 'Google Zero' Event Horizon With New Website Venture
A new profile by The Verge follows entrepreneur Campbell, who is launching a website business despite Google's zero-clic...
Anthropic's Claude Opus 4.8 Prioritizes Honesty: 4x Less Likely to Make Unsupported Claims
Anthropic released Claude Opus 4.8, claiming it is significantly more honest than previous models—early tests show a 4x ...
SOURCETRACKER: Hybrid Vector-Fingerprint System Enables Scalable Provenance Tracking for LLM-Generated Code
A new hybrid system combining vector search with Winnowing fingerprints achieves logarithmic-time provenance tracking fo...
Frontier LLMs Disagree on Fact-Checks: Study Reveals Reliability Gaps
A Hacker News discussion highlights a study showing leading AI models often contradict each other on real-world fact-che...
DeepMind Researchers Propose Cognitive Framework to Measure AGI Progress
A new arXiv paper from DeepMind and collaborators outlines a cognitive framework for measuring progress toward artificia...
Alipay Discloses 300M AI Transactions, Launches AI Wallet and Token Pay for Agent Economy
Alipay announced 300 million AI-powered transactions and support for 95% of general-purpose AI agents, unveiling an AI w...
AgentHijack Benchmark Exposes Fragility of Computer-Use AI Agents to Environment Corruptions
New benchmark from ICML 2026 reveals that state-of-the-art computer-use agents fail catastrophically under common UI cor...
Alipay Crosses 300 Million AI Transactions, Debuts AI Wallet and Token Pay Infrastructure
Alipay disclosed it has processed 300 million AI-powered payments and launched an AI wallet and Token Pay, claiming the ...
BestBlogs Launches AI-Powered Reading Assistant with Personalized Daily Briefs, Targeting Developer Audience
BestBlogs introduces AI-driven reading assistant that curates content from RSS, X, YouTube, and podcasts, offering a Pro...
Robin: World's First Fully Automated AI Scientist Completes Research in Two Hours, Report Says
AIbase.cn reports the launch of Robin, claimed to be the first fully automated AI scientist capable of completing a rese...
Memory Costs Surge to Dominate AI Chip Components, Epoch AI Analysis Reveals
Epoch AI reports memory now accounts for nearly two-thirds of AI chip component costs, up from roughly half in prior gen...
AI Voice Cloning Raises Ethical Questions as Tech Recreates Deceased Pilots' Voices
TechCrunch reports on a project using AI voice cloning to resurrect voices of dead pilots for training simulations. This...
Heart Rate Sensors in AirPods Pro 3 Signal Deeper Health AI Integration in Consumer Audio
The AirPods Pro 3 now include fairly accurate heart rate sensors, marking Apple's push into health monitoring via audio ...
Google AI Search Lures Users In, But at What Cost to the Web?
A new Wired analysis argues that Google's AI-crafted search results are so convenient that users will adopt them despite...
Google's Gemini for Science Marks a Pivot from Specialized AI to Agentic Research
At Google I/O, DeepMind's Demis Hassabis framed AI as entering the 'foothills of the singularity' while unveiling Gemini...
Study: More Capable LLMs Make Worse Forecasts When It Matters Most
A new arXiv preprint finds that as LLMs become more capable, their forecast accuracy degrades in high-stakes scenarios. ...
Claude Surges 130% Month-Over-Month in US Desktop AI Rankings, Closing Gap with ChatGPT
March 2026 US desktop AI rankings show Claude's usage skyrocketing 130% month-over-month, narrowing ChatGPT's lead. The ...
Gulf’s AI Boom Faces Undersea Cable Vulnerability, WIRED Reports
Hyperscalers are pushing Gulf states to overhaul internet infrastructure as AI data centers increase dependence on subse...
The Gulf’s AI Boom Faces an Undersea Cable Bottleneck, Wired Reports
Wired reports that hyperscalers are pressuring Gulf states to revamp undersea cable infrastructure as AI demand strains ...
AI Used to Resurrect Voices of Dead Pilots, Raising Ethical Questions
TechCrunch reports that AI voice cloning is recreating voices of deceased pilots for training and memorials. This practi...
Antigravity 2.0 Tops OpenSCAD 3D LLM Benchmark, Pushing AI-Generated Architectural Design
ModelRift's Antigravity 2.0 achieved the highest score on the OpenSCAD Architectural 3D LLM Benchmark, demonstrating sig...
Can OpenAI's 'Master of Disaster' Chris Lehane Fix AI's Reputation Crisis?
OpenAI global affairs chief Chris Lehane is moderating the AI risk debate while pushing states for favorable legislation...
xAI's Grok Appears in Only 3 of 400+ US Government AI Use Cases, Reuters Finds
A Reuters review found Grok was used in just 3 of over 400 federal AI records, signaling weak government adoption. This ...
Firefox's Project Nova Adds a One-Click AI Off Switch: A Browser Privacy First
Mozilla announces Project Nova, a major browser refactoring that introduces a single toggle to disable all AI features. ...
xAI's Grok Absent From US Government AI Adoption: Reuters Finds Only 3 Mentions
A Reuters review of over 400 government AI use cases found Grok appears in just three, all for basic tasks. The data rai...
Researchers Propose Full Attention Transfer to Sparse Models in Just 100 Training Steps
A new method from RTP-LLM shows sparse attention models can match full attention quality after only 100 training steps. ...