Back to today’s briefing

Story perspectives

AI Struggles with Analog Clocks, Reveals New Study

5/20/2025

31 4

Summary

Story summary

A groundbreaking study by researchers from China and Spain reveals that GPT-4.1 falters when reading analog clocks, frequently misreading time due to subtle visual differences. Even after fine-tuning on synthetic data, the model's accuracy plummeted with unfamiliar clock designs, raising alarms about AI's dependence on pattern recognition over genuine comprehension and the effectiveness of existing training techniques.

1 of 1

Story summary

A groundbreaking study by researchers from China and Spain reveals that GPT-4.1 falters when reading analog clocks, frequently misreading time due to subtle visual differences. Even after fine-tuning on synthetic data, the model's accuracy plummeted with unfamiliar clock designs, raising alarms about AI's dependence on pattern recognition over genuine comprehension and the effectiveness of existing training techniques.