When Apple’s AI started making up fake news: the bumpy ride of Apple in the world of AI
Apple, a company known for its near-flawless technological innovations, decided to dive into the world of artificial intelligence with a product called Apple Intelligence. But instead of leaving them in awe, it left them scratching their heads. Apple’s AI became so good at creating fake news and delivering bizarre analyses that the company had to hit the pause button on the entire program to avoid further embarrassment.
They Were Warned!
What really makes this story more interesting, however, is that even Apple’s own engineers had warned from the very beginning about the fundamental flaws in the technology. In fact, in a study released last October, they showed that these models have little to do with what we call “reasoning.” Instead, they merely mimic the patterns they’ve observed in their training data.
Math Is Too Hard for AI!
To test the AI models, the researchers challenged them with thousands of basic math problems. For example:
“James buys 5 packs of beef, each weighing 4 pounds. The price of beef is $5.50 per pound. How much did he pay?”
These problems were well within the capability of a middle schooler. But the researchers used a simple trick: they changed the numbers. This small change was enough to knock the models’ accuracy.
When they added irrelevant details—like mentioning that some fruits in a counting question were “smaller than usual”—things got even worse. The models’ performance plummeted catastrophically, with accuracy dropping by up to 65%. Even top-performing models like OpenAI’s o1-preview suffered a notable 17.5% drop.
AI or Copy Machine?
That experiment proved these models don’t actually “reason.” They are sort of high tech copying machines-great at an appearance of smarts while flailing miserably on minor changes to its problems.
Why Did Apple Push It Live?
Despite all the warnings and obvious flaws, Apple released its AI model. Maybe to keep up with the times or because that is just how it goes in the tech world: embracing risks even when the red flags are waving.
Conclusion
This story underlines the fact that even giants like Apple are not immune to challenges with AI, but it really shows how far we are from actually substituting human reasoning with artificial intelligence. Maybe for now, we need middle schoolers to solve our math problems!
Source: Futurism
