‘Complete collapse’: Bombshell blow to AI

Apple’s recent paper reveals fundamental limitations in advanced AI, showing that large reasoning models collapse under complex challenges while large language models perform better on simpler tasks but still fail on more demanding problems. Researchers noted that reasoning models reduce their effort when struggling, raising concerns about their reliability. Associate professor Niusha Shafiabady warned against treating AI as a magic wand, emphasizing the unpredictability of real‑world interactions. Gary Marcus criticized the hype, calling it a “bait and switch.” The findings also prompted Apple to suspend its AI‑powered news alerts after users reported serious accuracy errors, and a BBC study found significant misinformation across major AI assistants, including Google’s Gemini, OpenAI’s ChatGPT, and Microsoft’s CoPilot.