Jingnan Huo

Um, for example, you know, out of, out of this latest surge of interest, you know, in, in AI safety, both Anthropic and OpenAI have been talking about how, uh, they wanna have these third-party evaluators, um, sit in and evaluate, um, their products before they're released.

And so Huo Jingnan, who's one of the reporters on our team, you know, did a very, like, good and careful story kind of looking at, you know, who these evaluators are, and there are a couple of big issues with them, and one of which is the evaluators say they're not getting enough time to evaluate these models.

All the researchers I talk to say that evaluation results don't do a very good job of predicting model behavior.

They can predict all the ways peop- people will interact with AI models, and they can predict how AI models will behave in situations that evaluations didn't test for.

A parent has filed a class action lawsuit against OpenAI for allowing her teenage children to use ChatGPT without parental consent and profiting off their data.

NPR's Huo Jingnan has more.

The lawsuit, filed in California, alleges that OpenAI allowed the plaintiff's children to bypass its parental consent requirement easily and that the chatbot continued to engage with one of the children even after the girl said that she was twelve, below the minimum user age that OpenAI laid out in its own policy.

According to the complaint, the children shared intimate life details with the chatbot, including selfies, their anxieties about their appearance, and a photo of insects in the bathroom.

The plaintiff alleges that OpenAI unjustly profits off of such user data and demands compensation from the company.

In a response to NPR, OpenAI says it strongly disagrees with this, quote, "misguided lawsuit." It says it requires parental consent for underage users and lets teen users opt out of letting OpenAI train on their data.

Huo Jingnan, NPR News.

Americans have a new sunscreen option for the first time in over twenty-five years.