JD.com Pledges 22 Billion Yuan Housing Support for Couriers as China’s Instant Retail Competition Heats Up 
Azul Airlines Wins Court Approval for $2 Billion Debt Restructuring and New Capital Raise 
Intel’s Testing of China-Linked Chipmaking Tools Raises U.S. National Security Concerns 
Samsung SDI Secures Major LFP Battery Supply Deal in the U.S. 
Trello Outage Disrupts Users as Access Issues Hit Atlassian’s Work Management Platform 
Gulf Sovereign Funds Unite in Paramount–Skydance Bid for Warner Bros Discovery 
China Adds Domestic AI Chips to Government Procurement List as U.S. Considers Easing Nvidia Export Curbs 
CVS Health Signals Strong 2026 Profit Outlook Amid Turnaround Progress 
IBM Nears $11 Billion Deal to Acquire Confluent in Major AI and Data Push 
SpaceX Insider Share Sale Values Company Near $800 Billion Amid IPO Speculation 
Trump’s Approval of AI Chip Sales to China Triggers Bipartisan National Security Concerns 
SpaceX Reportedly Preparing Record-Breaking IPO Targeting $1.5 Trillion Valuation 
SK Hynix Shares Surge on Hopes for Upcoming ADR Issuance 
Apple App Store Injunction Largely Upheld as Appeals Court Rules on Epic Games Case 
Evercore Reaffirms Alphabet’s Search Dominance as AI Competition Intensifies 
Moore Threads Stock Slides After Risk Warning Despite 600% Surge Since IPO 
Australia Enforces World-First Social Media Age Limit as Global Regulation Looms 
Apple Inc. (NASDAQ: AAPL) has cast doubt on the reasoning abilities of today’s leading AI models in a new research paper titled “The Illusion of Thinking: Understanding the Strength and Limitations of Reasoning Models via the Lens of Problem Complexity.” The study evaluated large reasoning models (LRMs) such as OpenAI’s O1/o3, DeepSeek-R1, Claude 3.7 Sonnet Thinking, and Gemini Thinking, revealing significant performance declines as task complexity increased.
Using controlled algorithmic puzzle environments, Apple researchers demonstrated that these state-of-the-art models consistently failed to solve complex problems and lacked scalable reasoning capabilities. They noted that beyond a certain threshold of difficulty, the models' accuracy dropped to zero, exposing critical limitations in general problem-solving and adaptability.
The paper also criticized current AI evaluation benchmarks, suggesting they overestimate the true capabilities of modern LRMs. Apple instead proposed more rigorous testing environments to better assess how models handle abstract, non-standard tasks. Researchers concluded that despite their size, these models exhibit fundamental inefficiencies and cannot yet emulate the flexible reasoning seen in human cognition.
This research adds to growing skepticism about the proximity of general artificial intelligence (AGI)—a hypothetical form of AI capable of human-like understanding and reasoning. Current large language models primarily rely on pattern recognition and predictive algorithms, making them prone to logical errors and inconsistency in reasoning.
The paper’s release comes just ahead of Apple’s Worldwide Developers Conference (WWDC) 2025, where anticipation remains subdued amid criticism that the company has lagged rivals in AI development. Despite a partnership with OpenAI, Apple’s much-hyped “Apple Intelligence” features have faced delays, raising questions about its readiness to compete in the AI race.
This study underscores Apple’s critical view on the industry's AGI ambitions while signaling a renewed focus on foundational AI research.
