Photo of DeepakNess DeepakNess

The illusion of thinking

Unproofread notes

Apple published a new paper titled "The Illusion of Thinking: Understanding the Strengths and Limitations of Reasoning Models via the Lens of Problem Complexity" where they proved that AI reasoning models do actually reason.

I got to know about this from this post on X, where Ruben Hassid explains this is very detail.

It's claimed that after a point, no matter how much computing power you provide, these models can't solve harder problems that they haven't seen before. And as problems get harder, their thinking capabilities come lower.

The authors of the paper try to prove that models like Claude, DeepSeek, o3-mini, etc. do not "reason" at all.

You can access the paper directly from here as a PDF file format. I had a quick glance and it's an interesting read.

Comment via email