Apple researchers now raising red flags that may be hard for the AI trade to ignore.

In a new paper, they're throwing cold water on the growing trend of reasoning models and whether

they're really more accurate. Deirdre Bosa has more in today's tech check. Please explain, Deirdre.

I will. So in doing this, Apple is essentially calling out not just a growing, but the biggest

trend in AI right now.

And that is reasoning the ability of models to try and think like humans.

Now, these models are everywhere.

Open AI, Google, Anthropic, they are all racing to build AI that doesn't just spit out

answers, but shows its work step by step, promising more reliable and logical responses.

But Apple scientists, they're now saying that beyond a certain point, it just doesn't work.

Not only do these models struggle with harder problems, but they actually get worse as the

task gets more complex and they waste computing power overthinking simple tasks.

Here's an example from the paper.

Now, the AI, it's playing a simple version of

checkers, swapping pieces from one side to the other. It starts easy, but the more pieces you add,

the harder it gets. It's a logical puzzle. This chart shows how well AI models handle the puzzle

as it gets harder. At first, they're accurate, but at around eight to ten pieces, everything

just collapses. Even the most advanced reasoning models, they're accurate, but at around eight to ten pieces, everything just collapses.

Even the most advanced reasoning models, they break down.

So put simply, thinking harder doesn't make these models any smarter.

It just makes them slower, less reliable, and importantly more expensive, too.

They're burning through compute power.

...