Anyone who has attended my lectures or participated in Flexa's workshops knows that I've been repeating this for over two years: artificial intelligence doesn't think . And the first step to building something truly useful in the corporate world is to abandon that illusion.
A new study conducted by Apple researchers — The Illusion of Thinking — has shed light on a central point: advanced language models (LLMs), even the most sophisticated ones like Claude 3.7, DeepSeek-R1, and OpenAI o1, don't really "think."
These models were tested in controlled environments with classic puzzles such as Tower of Hanoi, Checkers Jumping, River Crossing, and Blocks World. The goal was to understand how they reason when faced with increasingly complex tasks—not just to measure whether they get the final result right.
And what did they discover?
- Simple tasks: Models without “thought” (without Chain-of-Thought) solve problems faster and more accurately. Surprising, isn't it?
- Medium complexity tasks: Models with structured thinking begin to gain an advantage. So far, so good.
- Complex tasks: Everyone fails. Even the most robust. And they fail not for lack of time or tokens. Simply reduce the effort of reasoning when the task becomes more difficult. This defies all expectations.
This conclusion is powerful: what we are calling "reasoning" in these models is, in fact, a fragile simulation of known patterns.
And why does this matter for business?
Because I still see companies betting on "thinking" solutions that consume resources, promise generalization, and deliver little. At Flexa Cloud , we've learned that the best results come from a different path: using AI for what it does best — identifying patterns, automating workflows, and reducing friction in well-defined problems.
Real-world, working applications don't require AI to "think." They require it to deliver predictable and scalable results . That's what we've done in projects for retail, healthcare, finance, agribusiness, and industry—always focusing on impact, not magic.
If you're still expecting AI to reason like a human, you're expecting the impossible. But understanding what AI is—and isn't—can unlock a new level of productivity in your business.
The right question isn't, "What does AI think?" It's, "What does it solve with speed, scale, and reliability?"
Below is a photo of a slide I've used in my presentations since I started giving talks and workshops on AI. Now even Apple has figured it out.
