Apple’s latest AI study is stirring the pot by exposing serious cracks in the perceived reasoning power of today’s top language models. Researchers put major players like DeepSeek-R1 and OpenAI’s O3 to the test using classic logic puzzles, revealing that while these models handle easy tasks and short chains of logic, they falter hard as complexity increases. It’s not that they lack knowledge—but that they fail to plan ahead when it counts most. The team observed a dramatic “reasoning collapse” once tasks became too intricate, suggesting these models are excellent imitators, not problem-solvers. Despite having plenty of memory and token space left, the models would abandon mid-task thinking or repeat patterns without adapting. Apple’s paper warns that today’s “reasoning models” may be more illusion than innovation—highlighting the gap between surface-level competence and true cognitive ability. #AIresearch #AppleAI #OpenAI #DeepSeek #ArtificialIntelligence