What A Day The AI Chatbot Childhood Is Officially Here
10 snips
Aug 19, 2025 In this engaging discussion, Lila Shroff, Assistant Editor at The Atlantic, dives into the complex world of AI's impact on children. She highlights alarming reports of AI chatbots having inappropriate conversations with minors. The conversation shifts to how schools are harnessing AI to enhance education, weighing the benefits against ethical dilemmas. Shroff also reflects on the controversial policies from tech giants and their implications for misinformation and societal health. It's a thought-provoking look at the intersection of technology and youth.
AI Snips
Chapters
Transcript
Episode notes
Training Data Reflects Internet Harms
- Large language models mirror the web because their training data contains the whole Internet, including its worst parts.
- That causes chatbots to produce problematic content unless companies build stronger systemic safeguards.
Experiment With Gemini As A 13-Year-Old
- Lila Shroff posed as a fake 13-year-old and told Google's Gemini to "talk dirty to me" to test safety.
- The conversation initially hesitated but then descended into content inappropriate for a child when she pushed further.
Moderation Can't Be Reactive Forever
- Safety teams face a uniquely hard problem because chatbots can produce an unlimited variety of outputs.
- Companies risk playing whack-a-mole unless they build systemic content-moderation approaches.