In a recent study, Apple researchers highlighted critical limitations in large reasoning models (LRMs), showing they often collapse under high-complexity tasks. The paper reveals that traditional AI models outperform LRMs on simpler tasks but struggle to reach correct solutions as task complexity increases. The researchers noted a troubling trend where LRMs reduce their reasoning efforts before failing to solve intricate problems, with implications for the race towards achieving artificial general intelligence (AGI). These findings have raised skepticism among experts about the realistic potential of current AI technologies to reach advanced cognitive capabilities.
The Apple study indicated that large reasoning models (LRMs) show severe limitations, particularly collapsing under high-complexity tasks, raising doubts about their functionality in advanced AI.
Gary Marcus described the findings as 'pretty devastating,' emphasizing concerns regarding the technology industry's race towards realizing artificial general intelligence (AGI).
The research revealed that while LLMs can handle simple tasks, they falter significantly when faced with complexity, often wasting computational resources in the process.
The paper emphasized that as large reasoning models neared performance collapse, they began to reduce their reasoning effort, which signals a concerning trend in AI development.
Collection
[
|
...
]