ChatGPT's invisible watermarks

This link leads to one of the most interesting things I’ve seen in AI recently.

What’s cookin’? OpenAI is rolling out invisible watermarks for some ChatGPT and Codex text in the EU to comply with new AI transparency rules. The system, called textGrain, doesn’t add a visible label. Instead, it subtly changes which words the model chooses, creating a statistical pattern that special detectors can recognize. OpenAI says the watermark does not noticeably hurt model performance, but it is far from foolproof. In testing, replacing just 10% of a passage with synonyms dropped detection from about 92% to 66%. Shorter, heavily edited, translated, or highly technical text is also harder to identify.

🤔 Hunter’s take: This is interesting because it shows how messy “AI detection” really is. A watermark can suggest AI was involved, but it cannot tell you how much a human contributed, whether the writing is accurate, or even prove that unmarked text was human-written. Ultimately, watermarking can be useful as a clue, but not as a verdict.

What’s cookin’? A new study found that patients were less likely to trust, choose, or follow the advice of doctors who disclosed that they use AI. Researchers showed more than 1,000 U.S. adults profiles of family doctors, with some saying they used AI and others saying they never did. The doctors who used AI were rated as less warm and less competent, and people were more likely to seek a second opinion, travel shorter distances to see them, and pay less for an appointment. Interestingly, the effect was strongest among adults under 45.

🤔 Hunter’s take: This is a fascinating trust problem, especially in younger people who grew up with more advanced technology. The lesson for doctors probably isn’t “don’t use AI.” It’s “don’t make AI sound like it replaced your judgment.” Patients seem to care a lot about knowing a human is still responsible for the decision, and they don’t (yet) trust AI with their health.

Here’s your chance to get a straightforward answer, from an expert in the field.

Hi — I’m Hunter, a PhD candidate whose work has appeared in major academic journals and popular tech outlets. I founded FryAI to make staying ahead of AI clear, accessible, and fun.