Story perspectives

Study Reveals Junk Data Harms AI Reasoning Abilities

10/22/2025

29 2

Summary

Story summary

Researchers from Texas A&M University, the University of Texas at Austin, and Purdue University test the LLM Brain Rot Hypothesis, arguing exposure to low-quality web text can cause cognitive decline in large language models.
They ran experiments on four LLMs and found junk data significantly impaired reasoning.
The decline followed a dose-response pattern: accuracy fell from 74.9% to 57.2% as junk data rose; popularity, not semantic quality, predicted decline.

1 of 1

Story summary

Researchers from Texas A&M University, the University of Texas at Austin, and Purdue University test the LLM Brain Rot Hypothesis, arguing exposure to low-quality web text can cause cognitive decline in large language models.
They ran experiments on four LLMs and found junk data significantly impaired reasoning.
The decline followed a dose-response pattern: accuracy fell from 74.9% to 57.2% as junk data rose; popularity, not semantic quality, predicted decline.