Siri, are you okay? Apple’s AI gets a reality check

Tech giant Apple has published a research paper that effectively side-eyes its own artificial intelligence.

The report, The Illusion of Thinking, finds that Apple’s AI models, the same ones powering the newly branded Apple Intelligence suite, falter under pressure.

When the tasks get tough, the models often flub the answer or simply appear to stop trying altogether.

In tests conducted by Apple itself, the company’s flagship models underperformed when stacked against competitors from OpenAI, Google, Meta, and even Alibaba.

That includes both the Apple On-Device model, which runs directly on hardware like iPhones, and the more powerful Apple Server version, which handles processing in the cloud.

Industry implications

The findings have elicited strong reactions from the AI community.

AI expert Gary Marcus described the report as "devastating,"  in his Substack, suggesting that current models may not be the direct path to transformative artificial general intelligence (AGI) .

Marcus also added that anyone who believes current models are the fast track to truly transformative intelligence is “kidding themselves.”

Apple's cautious approach to AI development, emphasising privacy and on-device processing, contrasts with the more aggressive strategies of competitors like OpenAI and Google.

Real-world applications

It comes at a time when Apple has faced public missteps in its AI rollout, including the recent suspension of its iPhone news alerts after a series of fact-check fails.

Those included a false alert about Rafael Nadal, a wrongly reported death, and a premature darts championship win.

But Apple isn’t alone in these blunders, with other major players like Google, OpenAI, and Microsoft also grappling with AI-generated misinformation.

Keep on top of the most important media, marketing, and agency news each day with the Mediaweek Morning Report – delivered for free every morning to your inbox.