Apple Blasts Reasoning Models as "Fake Thinking"!
Apple's AI paper challenges large models' reasoning, calling it 'pattern matching'. Tests show collapse in high-complexity tasks.
"AI Disruption" Publication 6800 Subscriptions 20% Discount Offer Link.
Apple’s latest large model paper has caused a stir in the AI community.
Some have summarized it as: Apple just played the role of Marcus, denying the reasoning capabilities of all large models.
The paper claims that reasoning models are not truly thinking, whether it’s DeepSeek, o3-mini, or Claude 3.7—they are all just another form of “pattern matching”, and the so-called thinking is merely an illusion.
When faced with truly high-complexity tasks, all models collapse, even if given sufficient time and computational resources.
The authors include Samy Bengio (brother of Turing Award winner Yoshua Bengio), one of the founders of Google Brain.
Some netizens sarcastically remarked that despite Apple’s vast financial resources, it has failed to produce notable results in two years, and now, lagging behind, it’s criticizing others’ achievements.
Others suggested that Apple might as well buy Anthropic, the company behind Claude, as the cost keeps rising with every passing day.
However, some pointed out that the paper isn’t as negative as it seems, but rather calls for establishing better reasoning mechanisms and evaluation methods.
So, what exactly does this paper say?