Apple’s ‘The Illusion of Thinking’ is shocking – but here’s what it missed

Apple just said, “Allllllll y’all reasoning models are full of…” 💩

Fine, it didn’t actually say that. 🤗
That was my loose interpretation of Apple’s recently published report, “The Illusion of Thinking.”

Also: Is ChatGPT Plus really worth $20 when the free version offers so many premium features?

If you aren’t familiar, Apple just confirmed what many suspected for the longest: there’s no “thinking” happening in Large Reasoning Models (LRM); it’s all advanced pattern matching. 😱

This report brings up soooooooo many questions. To start, are big corporations using AI as a scapegoat to cut jobs and thus increase profits?

Or could “lack of intelligence” be the reason “Big AI” is making a push to hardware and robotics to keep the hype train going?

SO MANY QUESTIONS, SO LITTLE TIME!

In today’s article, I will bring you up to speed on what Apple reported, why it matters to you, and what your next steps should be.

I can guarantee it’s not what you think. 🤓☝️

If you’re new to my work, my name is Lester, but feel free to call me Les. I’m a founder with a successful exit, currently the executive chairman of a group of ecom brands, and an award-winning performance marketer.

Also: How ChatGPT could replace the internet as we know it

My team has adopted AI to develop internal tools that allow us to stay ahead in a highly competitive space, which gives me real insight into how AI is transforming digital marketing and the opportunities that come with it.

If you’re finding it hard to stay ahead with AI and want insights and opportunities in this crowded space, considersigning up for my free newsletter, No Fluff Just Facts. I share what’s working, what to watch for, and the AI innovations that are worth your time.

But enough about me. I have to tell you all the tea on “The Illusion of Thinking”… you won’t believe how many people are missing the bigger picture. 😉

The illusion of thinking

Before we jump in, let’s define what a reasoning model is. A Large Reasoning Model (LRM) is an AI system designed to solve complex problems by working through each step, essentially showing its reasoning process before providing an answer. 🧐

Unlike standard language models that predict the next word, these models are trained to apply structured reasoning and logic, breaking down problems in a more human-like, step-by-step way.

Also: I told ChatGPT more about myself – here’s how the AI used that personal info

Apple’s “The Illusion of Thinking” put some of the top Large Reasoning Models like OpenAI’s o3 mini, DeepSeek R1, and Claude 3.7 Sonnet to the test with four classic puzzles: 🎮

• Tower of Hanoi
• Checkers Jumping
• River Crossing
• Blocks World

As the tests became harder, the models hit a wall like Wile E. Coyote. 💥

Womp Womp Wompppppppppppppppppp.

The models’ performance decreased as complexity increased, and adding more computing power or tokens did not help.

The more complex the puzzles became, the less these models thought, even though they had the capacity to continue. They basically gave up — just like humans! 😩

Anyway, Apple broke down the results into three performance zones:

• Low complexity: Standard models did better and used fewer tokens.
• Medium complexity: Thinking models showed an advantage.
• High complexity: All models failed. Accuracy dropped sharply.

Even when Apple gave the LRMs the answers to the test in the form of an algorithm, the models still failed and couldn’t follow a multi-step plan to completion.

The LRMs’ ability to reason was inconsistent. A model could complete over 100 moves in Tower of Hanoi, but could only manage fewer than 5 correct moves in River Crossing.

Also: This new ChatGPT trick lets you download your Canvas work as a PDF in seconds – here’s how

This behavior suggests pattern matching, not true reasoning or intelligence. 😓

All that said, “The Illusion of Thinking” started a turf war and was met with a good amount of criticism. It’s important to note that this paper is a preprint and has not undergone peer review.

In other words, Apple put out the paper and was like, “Trust me, bro.”

Another criticism making the rounds online is that Apple used puzzles rather than real-world scenarios. Critics also argue that the puzzles exceed the models’ token and step limits, so the “failures” reflect these constraints rather than a lack of reasoning.

But the most hilarious critique has to be, “Apple is behind on AI, so it’s pointing out the weak spots of rival systems after losing ground.”

Completely random, but when I read that critique, I can’t help but think of Denzel Washington in “Training Day.” 🍿🎥

“King Kong ain’t got ish on me!”

Apple’s probably too classy for such colorful language, but the thought cracks me up nonetheless. Hehehe.

Movie quotes aside, let’s discuss your next move. 💁‍♂️

Your next move

Let’s get serious for a second. 

Apple did the evaluation, taking its ball and bat and going home.

If true, good on Apple for calling out the AI equivalent of “natural flavors,” exposing that one of the most significant AI advancements might just be a product of the marketing department.

Also: Why I’m switching to VS Code. Hint: It’s all about AI tool integration

Meanwhile, the internet is left to slug it out. One camp claims AI is all hype, while the other argues the test was rigged from the start.

Do you buy into the hype, or do you ignore AI altogether? 🤔

Here’s the thing: if you’ve ever used AI in a meaningful way, you know there’s some truth to what this report revealed. Even if Apple pushed it a bit too hard, it doesn’t change the fact that, based on our interactions, we recognize that AI is still lacking and has a long way to go.

Currently, AI performs best when we’re the subject matter experts and can identify its mistakes.

I don’t think this report changes the bigger picture. In fact, it kinda doesn’t matter. 😤

Reasoning models might be more hype than substance right now, but that doesn’t mean AI won’t take jobs or transform industries. Dismissing AI today would be like dismissing the early computer just because it was slow.

Even if AI doesn’t live up to the brochure, its potential is incredible. Speaking for myself, AI has enabled me to brainstorm effectively and automate tedious tasks, ultimately giving me a level of productivity that would require working 26-hour days.

Also: The best AI for coding in 2025 (and what not to use)

While it’s not perfect and gets on my last nerve at times, it empowers everyday people like you and me to be more independent and creative when working on our businesses or careers.

So, what should you do? 😫

First, continue to stay informed, but also implement what you learn. Moreover, don’t get too high on the highs or too low on the lows when it comes to AI.

Outside of that, here are a few more recommendations:

• Use AI as a tool, not a crutch: Think “do it with me,” not “do it for me.”

• Don’t dismiss AI: Just because it’s flawed doesn’t mean it can’t be helpful.

• Be the subject matter expert: Continue to learn and grow in your field. The more you know, the more useful AI becomes.

• Automate boring stuff: Free up your time by letting AI handle the repetitive tasks so you can continue to learn and implement.

• Double down on human strengths: AI is powerful when combined with human capabilities such as leadership, empathy, problem-solving, and collaboration.

• Take action, even if it’s messy: Don’t wait for AI to be perfect or for the perfect prompt. Get off the sidelines and start playing with AI if you haven’t already. The people winning with AI are those who are using it.

Lastly, this one serves as a valuable life lesson: be mindful of potential agendas. My dad once told me, “A carpenter will always tell you to build a house using wood.” 🤓☝️

My two cents

As I step down from my soapbox, I want to make something abundantly clear…

I’m not knocking Apple’s report. I believe Apple’s findings are spot on based on my interactions with these models, but so what? ✅

Jobs will still be lost, and industries will continue to be transformed by AI. That’s just a fact.

While the report is insightful, it doesn’t fundamentally change the reality we’re facing. At the same time, I’m not here to defend LRMs either.

If anything, I’m on “Team You,” and as your teammate, I’m here to remind you: don’t throw out the baby with the bathwater.

😇 Hope this helps. I’m rooting for you.

P.S. If you want more easy and helpful AI tips and tricks, sign up for my free newsletter,No Fluff Just Facts.