From Cointelegraph

June 9, 2025 12:27 AM:

The race to develop AGI is ongoing, as Apple researchers find that leading AI models struggle with reasoning. Despite recent updates, the capabilities and limitations of LRMs in AI models remain poorly understood. Current evaluations focus on final answer accuracy, not reasoning abilities, challenging the expectation of AGI being near. Apple researchers test AI models’ thinking abilities through puzzle games, revealing limitations in exact computation and reasoning inconsistencies. LRMs fail to use explicit algorithms and reason inconsistently across puzzles, falling short of AGI-level reasoning. AGI is the ultimate goal of AI development, aiming to create machines that can think and reason like humans. OpenAI and Anthropic CEOs expressed confidence in building AGI, with expectations of surpassing human capabilities by 2026 or 2027. Despite recent advancements, LRMs in AI models still face challenges in generalizable reasoning, raising questions about current approaches to achieving AGI.

Read more at CoinTelegraph: AI Models Lack Reasoning Capability Needed For AGI