OpenAI enhances safety, rolls out age prediction measures in ChatGPT

What you need to know

• OpenAI announced that it's rolling out "Age Predictions" in ChatGPT for users in the EU.
• Age Predictions will consider several account-level aspects to make a judgement about the user's age, and if they are deemed under 18, it will impose restrictions to limit sensitive content.
• Last week, OpenAI announced that ads are headed to ChatGPT, as well as its cheaper ChatGPT Go tier.

OpenAI is sticking to its word about bringing additional safety measures to its AI platform this week by "predicting" the user's age.

The company announced in a blog post that it's preparing to roll out "Age Predictions," which reportedly builds upon its existing safety measures. This parameter isn't foolproof, but that's secondary to OpenAI's confirmation that Age Prediction will "help estimate whether an account likely belongs to someone under 18." Your "stated age" is taken into account by the AI; however, OpenAI states its model will also consider the time you're typically active, as well as "usage patterns."

Now, what is all of this for? If ChatGPT determines that the user is (potentially) under the age of 18, it will reduce that person's exposure to sensitive content. Such content includes graphic violence, viral challenges that could pose a health risk, depictions of self-harm, and more.

ChatGPT will also limit content concerning "extreme beauty standards, unhealthy dieting, or body shaming." The innerworkings of ChatGPT's Age Prediction is similar to what Google did on YouTube.

OpenAI acknowledges the trials here, stating that it will utilize Age Prediction to refine the signals that influence its decision-making and minimize mistakes. If ChatGPT wrongly restricts your account, OpenAI states users can quickly verify their age by submitting a "selfie through Persona, a secure identity-verification service." On a related note, the update highlights what parents can do for teen accounts.

Improving AI safety

OpenAI says that parents will find controls to set "quiet hours" in ChatGPT, creating a window of opportunity for their teen to access the AI. Outside of those times, the AI is inaccessible. Additionally, parents can control what the AI "remembers" about their child, alongside tools to control its model training. OpenAI also says parents can opt into notifications if "signs of acute distress" are discovered.

The post adds that OpenAI will take things slow with its Age Prediction for accuracy purposes. This feature is rolling out in the EU "in the coming weeks" for "regional requirements."

OpenAI had a busy week last week, detailing the arrival of its cheapest ChatGPT Go tier. This subscription is meant to give people all the tools they need in the U.S. after observing a "high adoption rate" in other markets. With ChatGPT Go, users have a much higher usage cap than its free tier. Sending messages, uploading files, images, and more becomes a lot easier.

The other main highlight from last week's announcement was ads. OpenAI confirmed that ads are headed to ChatGPT's chats; however, it's more of a test. The post said that ads are designed to be "relevant" to your current conversation with the AI. OpenAI claims it won't use the conversations to influence the ads it shows.