AI 行业动态、深度分析与工具指南

Top AI Models Fail 25% of Fact-Check Questions Without Search

New data reveals significant hallucination rates in precision tasks

Analysis of 6,000 fact-based questions shows even leading AI models err on one in four answers when lacking real-time search capabilities.

The Loneliness Epidemic: How AI and Screens Erode Empathy

Experts warn that digital addiction and remote work are atrophying human connection.

As AI automates coding and social media thrives on dopamine loops, society risks becoming a collection of lonely, self-absorbed individuals losing the art of human empathy.

Playing to Spot AI Slop: New 'Slop Sense' Game Tests Human Instincts

Interactive challenges urge users to trust their eyes in an era of synthetic media.

A new suite of quick games called 'Slop Sense' challenges players to distinguish real photos from AI-generated images, testing human intuition against deepfakes.

Rowan brings static security scanning to AI applications

Open-source SAST tool scans code, models, and agent configs without executing them

Hedgerow Dev releases Rowan, a static security scanner for AI/ML projects that analyzes source code, model files, and MCP configurations with traceable evidence.

Labelbox Unveils Recursion to Automate Enterprise AI Workflows

Frontier AI platform lets agents work autonomously in the background.

Labelbox has launched Recursion, a new platform enabling enterprises to deploy AI agents that execute complex business tasks autonomously, from coding to financial analysis.

SEO for AI Agents: Can Search Bias Influence Tech Choices?

New research reveals that 'hacked' search results can boost AI agent recommendations for targeted cloud products by nearly 80%, signaling a shift from traditional SEO.

Apple to Restrict Full Disk Access on Macs Amid AI Risks

New macOS controls demand explicit user consent for sweeping file permissions.

Apple is tightening Mac permissions, requiring very explicit action for full disk access to mitigate threats from increasingly autonomous AI agents.

ThreadShelf Brings Unified Semantic Search to AI Chat Archives

Local-first tool aggregates conversations from six major providers into one searchable workspace.

ThreadShelf lets users archive, semantically search, and continue AI conversations from ChatGPT, Claude, Gemini, Grok, OpenRouter, and LM Studio in a single local workspace.

AI Workload Placement: Decision Engine Reveals Cloud Speed Advantage

A new decision engine for AI workloads shows cloud inference can beat local execution on latency, shifting the debate toward privacy and offline capability tradeoffs.

Meta Opens Source Muse Code for DIY Hardware Projects

Build your own Muse AI gadgets with the new open-source SDKs.

Meta has open-sourced its Muse AI agent code, allowing users to build custom hardware gadgets using Raspberry Pi, ESP32 boards, and various displays.

Sean Parker Pivots Stability AI to Music with $76M Label Backing

Stability AI is rebuilding around music professionals after securing $76 million from Sony, Warner, and Universal, marking a strategic pivot under Sean Parker.

Sling TV Retires One-Day Cable Passes

Dish-owned Sling TV has discontinued its day-pass feature, ending the option to buy single days of cable without a full subscription. The move follows recent legal challenges from major broadcasters.

How AI Security Tools Uncovered Hundreds of MariaDB Bugs

MariaDB attributes a record spike in security reports to AI-assisted research rather than new vulnerabilities.

AI-powered bug hunters flooded MariaDB with 81+ reports last quarter, proving that old code deserves fresh eyes and better tools.

BMad Method Gains Momentum as Agile AI Development Framework

Open-source tool helps developers integrate AI without losing strategic context

The BMad Method repository has surpassed 53,000 stars, offering an agile approach to AI-driven development that keeps decision-making explicit and context durable across coding workflows.

OpenAI’s Dot Agent: Enterprise Power Meets Personal Tasks

OpenAI's new AI agent blends workplace software with personal automation, though teething issues remain.

OpenAI launched Dots, an enterprise-grade AI agent available to Pro subscribers. Testing reveals strong capabilities for technical workflows but mixed results for routine personal tasks like ordering food.

Apple Tightens macOS Full Disk Access for AI Agents

New controls aim to curb risks as AI software gains deeper system access.

Apple is restricting macOS Full Disk Access to prevent AI agents from silently reading private data, following incidents involving Meta's Muse and ChatGPT vulnerabilities.

Chatham Cuts Trade Validation to Under 4 Minutes With OpenAI

Finance firm scales capital markets expertise using Codex and GPT-5.6

Chatham Financial has slashed trade validation time from 30 minutes to under 4 by integrating OpenAI's Codex and GPT models into its workflow redesign.

White House Rebrands AI as 'Super Intelligence' in New Executive Order

The White House has officially rebranded AI as 'super intelligence' via executive order, coinciding with a major safety pledge signed by top tech CEOs.

Consumer AI Adoption Stalls at 2% Despite White House Pledge

White House rebrands AI as 'super intelligence,' but the economics of consumer adoption remain stubbornly bleak.

Despite a new White House pledge signed by top tech CEOs and a federal rebrand to 'super intelligence,' only 2% of consumers are buying AI products.