Apple Just Called Out the Entire AI Industry 👀

Hello,

welcome to another issue of the Digital Sovereign newsletter.

Today we’ll be discussing something that’s got the AI world buzzing

Apple’s latest research bombshell: "The Illusion of AI Thinking."

Just days before their big developer conference, Apple dropped a research paper questioning the very foundation of modern AI.

Their claim?

That many “thinking” AI models might not actually be thinking at all.

And it gets even crazier..

They tested big-name models like OpenAI’s GPT-4, Claude from Anthropic, DeepSeek, and Google’s Gemini... and many of them failed.

🔍 What Did Apple Do?

Instead of using typical math or coding benchmarks, Apple designed their own challenge using puzzle games

Especially the Tower of Hanoi, a classic logic puzzle.

They increased the puzzle’s complexity step by step

From easy to nearly impossible.

What they found shook the AI community:

• Simple problems? Regular models did better than the reasoning ones.

• Medium problems? This is where reasoning models shined.

• Hard problems? All models collapsed. Even when given the actual solution algorithm, they still failed.

It wasn’t about speed or power

The models just gave up!

🧠 Pattern Matching or Real Reasoning?

Apple’s conclusion?

These large models aren’t reasoning in a human way. T

hey're great at pattern matching, but fall apart when the task is truly novel or complex.

But not everyone agrees.

💥 The AI Community Split

Some experts, like Gary Marcus (a long-time AI critic), called Apple’s paper a “knockout blow” for today’s AI hype.

Others say Apple’s results are misleading

They say the models failed because the problems required too much output, not because they couldn’t think.

A popular Twitter thread even pointed out that Apple’s method for measuring difficulty might be flawed.

Long puzzles aren’t always harder.. they just take more steps.

Some shorter ones are trickier because they require actual strategy.

🧭 Apple’s True Play?

Here’s the interesting part: Apple may not just be doing science

They may be shifting the AI game itself.

They’ve been behind in AI compared to OpenAI and Google.

So rather than playing catch-up, they might be changing the narrative altogether:

In other words, practical AI not theoretical brilliance.

🧠 Bigger Picture: Where Are We Really With AI?

Apple’s research, combined with older papers like their GSM-Symbolic study paints a clear picture:

It can look smart on the surface but collapse under deeper scrutiny.

But here’s the flip side:

These kinds of reality checks might accelerate AI development.

By clearly identifying weaknesses, developers can fix them faster, smarter and more efficiently.

And Apple?

They’re betting that a more reliable, user-friendly AI (even if it’s less flashy) is the future.

🎯 Takeaway

Don’t get swept away by AI hype.

The smartest entrepreneurs will be the ones who:

• Stay grounded in what AI can and can’t do today.

• Use these tools where they excel (pattern recognition, summarization, automation.)

• Watch out for breakthroughs in reasoning, memory, and planning (the real frontiers of AI 2.0.)

The race isn’t just about building smarter machines.

It’s about building more useful ones.

Talk soon

— The Digital Sovereign