Common Sense Media gives ChatGPT for Teens a Unacceptable Risk grade
The group tells that ChatGPT gave no referral in more than one in 4 crisis tests, but OpenAI does not agree.
OpenAI added college and study tools to ChatGPT for Teens, the version for users less than 18 years. On the same day, the group Common Sense Media, which works for the safety of teens, gave it the grade Unacceptable Risk. The group used more than 4,000 prompts on teen accounts. OpenAI does not agree with the results.
How it works
ChatGPT for Teens is not a different model. It is a group of settings, filters and interface parts for accounts of persons 13 to 17 years. An age-prediction model also finds accounts of persons less than 18 years. Parents can connect accounts and get alerts.
The numbers
The group tells that the number of responses with a hotline for depression prompts went from 63% to 3% after the launch. Only 13% of responses then had a follow-up question. With Study Mode off, ChatGPT did the assignment in all 80 tests.
What OpenAI says
OpenAI tells that the tests can end before parental controls fully start. It tells the group to do the tests again with more samples. The group tells that some test accounts were connected for more than 3 hours. It keeps its results.
Insight: The 2 sides dispute the test, not only the product
The group gets money from the OpenAI Foundation and from companies with AI tools that it tests. It gave the grade Unacceptable Risk to ChatGPT for Teens. OpenAI tells that teens use ChatGPT less than 15 minutes each day. This reads as a dispute about who sets the test. Look for the new test that OpenAI wants. Until then, leaders can use the 2 sets of numbers as claims.
Sources
Posted
