The Breakdown: Understanding How Your Data Fuels AI Training and Why Chatbots Aren’t Medical Professionals

Sure! Here’s the rewritten content with the requested adjustments:

Millions of images of passports, credit cards, birth certificates, and other documents containing personally identifiable information are likely included in one of the largest open-source AI training sets, new research has found. Thousands of images—including identifiable faces—were discovered in a small subset of DataComp CommonPool, a significant AI training set for image generation scraped from the web. Since the researchers audited just 0.1% of CommonPool’s data, they estimate that the actual number of images containing personally identifiable information, including faces and identity documents, is in the hundreds of millions. 

The bottom line? Anything you put online can be, and likely has been, scraped. 

—Eileen Guo

AI companies have ceased to consistently include warnings that their chatbots aren’t medical professionals. New research indicates that AI models now often respond to health questions directly, sometimes even providing follow-ups and attempting diagnoses. These disclaimers are vital reminders for individuals inquiring about serious topics, such as eating disorders or cancer diagnoses. Their absence can lead users to trust potentially unsafe medical advice.

—James O’Donnell