Apple is turning heads with its latest research. But have they hit the mark or lost the edge in technology?
WWDC 2025 started with a bang and got people talking- everything from the infamous Liquid Glass to the fact that developers will now be able to run Linux containers on the Mac as VMs.
But for some reason, the WWDC was missing the enthusiasm behind its AI, Apple Intelligence- whose abbreviation is also AI- smart move, Apple.
Following the event, Apple released a paper called “The Illusion of Thinking.” In it, Apple’s researchers made the AI systems play various puzzles and tested their efficiency. Their conclusion was that AI models are just good at randomized outputs and probabilistic guesswork- not real thinking.
This stirred people up, and since Apple is known for pushing technological boundaries and having authority in the tech world- it stirred people up whose investments matter to the organizations creating these AI LLM and LRM models.
The paper is one of the most talked about things on the internet. However, A mathematician called Dr. Trefor Bazett posted his summary of the paper, and in it, he posits two observations (around the 6:53 mark)
- AI systems are not calculators.
- The number of tokens used was pushed beyond their maximum limit.
Anthropic and Open Philanthropy dropped a paper (The Illusion of Illusion of Thinking) in response- it said the same thing. They critiqued Apple for giving the LLMs problems that were unsolvable. And that they drove problems to the collapse of the token.
Fair criticism. They argue that the fundamental mode of evaluation was flawed.
Apple’s paper has some points.
- Why did the models exhibit thought failure before solving the problem?
- Why do models “overthink” in easy tasks but not generate meaningful C-o-T in complex ones?
- Why does model accuracy drop and not scale at complexity?
These are valid concerns. But Apple has been on the defense since their AI has failed to take off. The AI race has not been kind to the company.
The question that seems to have arisen is if that matters at all.
Maybe all our machines will stay pattern-recognizers and probability text machines incapable of true creativity.
Or we find a way to create sentience.