Archive · 9 Jun 2025
Apple Study: AI Reasoning Models Memorise, Don't Reason
Apple conducted a study that challenges the idea that current AI models can truly 'think'.
The study found that these models are good at recognizing patterns but struggle with complex problems.
They tested the AI on puzzles, increasing the difficulty to see if the AI would improve.
But the models' performance decreased significantly as the puzzles became harder.
This suggests that AI may not be as advanced as some think, and human-like AI could be further away than anticipated.
Apple published a study questioning the reasoning abilities of advanced AI models.
The study found that models excel at pattern recognition but fail with complex tasks.
Researchers tested models on puzzles like Tower of Hanoi, varying the difficulty.
Performance of the models decreased as puzzle complexity increased.
The findings suggest that Artificial General Intelligence (AGI) may be further away than some predictions.
- Who
- Apple researchers studied AI reasoning models like Claude, DeepSeek-R1, and o3-mini.
- What
- Apple conducted a study challenging the reasoning capabilities of advanced AI models.
- When
- The study was recently published.
- Where
- Not explicitly mentioned in the article.
- Why
- To assess the true reasoning abilities of current AI models and their potential for Artificial General Intelligence (AGI).
Current AI Capabilities
Future AI Potential
Reasoning Abilities
Current AI Capabilities
AI models can appear intelligent by recognizing patterns.
Future AI Potential
AI might not be capable of advanced reasoning like humans.
AGI Timeline
Current AI Capabilities
AGI may be achievable soon.
Future AI Potential
AGI, with human-like capabilities, may be further away.
- Company
- Apple
- Study Title
- The Illusion of Thinking
- Models Tested
- Claude, DeepSeek-R1, o3-mini
- Models' Strength
- Pattern Recognition
- Puzzle Games Used
- Tower of Hanoi, Checker Jumping, River Crossing, Blocks World