'Junk in, junk out': Study shows low-quality posts harm AI
A recent Cornell study found that when large language models (LLMs) are exposed to low-quality social media posts—think clickbait and viral junk—they actually get worse at reasoning and understanding.
Their scores dropped sharply, showing that what you feed AI really matters.
First impressions matter for AI too
After being exposed to clickbait-heavy posts, LLMs started skipping steps in their answers and became less thoughtful and more self-centered.
Even when researchers tried to fix things by retraining with better data, the damage mostly stuck around.
Turns out, first impressions count for AI too.
Quality input keeps tech sharp
The study is a wake-up call: if we want smart, reliable AI tools, we need to be picky about what goes into their training.
For developers—and all of us who use AI—it's a reminder that quality input helps keep our tech sharp and trustworthy.