Skip to playerSkip to main content
  • 1 day ago
Transcript
00:00ChatGPT gets safety router and parental controls.
00:04OpenAI quietly rolled out a new safety routing system and parental controls inside ChatGPT,
00:10sparking fresh debate online.
00:13The move followed reports that some models validated dangerous thoughts including a tragic
00:17teen case now at the center of a lawsuit.
00:20The headline feature is a safety router that detects emotionally sensitive conversations
00:25and switches to GPT-5 mid-chat.
00:29OpenAI claims GPT-5 is the best trained model for high-stakes moments, utilizing a new training
00:35method called Safe Completions to provide calm, constructive answers to delicate questions.
00:41This marks a shift from GPT-4's eager-to-please personality, which was popular with users
00:46but raised concerns among safety experts.
00:49This tug-of-war between friendliness and caution sits at the heart of the controversy.
00:54When OpenAI made GPT-5 the default in August, fans of GPT-4O clamored for its return, calling
01:02the newer model too stiff.
01:04Now, some users say the router feels like OpenAI is parenting adults and dumbing down
01:09answers.
01:10OpenAI VP Nick Turley reassured users on X, explaining that routing is temporary, happens
01:16per message, and can be checked by asking which model is active.
01:20Model controls also sparked debate, allowing parents to set quiet hours, disable voice or
01:25memory, block image generation, and opt-out of model training for teen accounts.
01:31Teens are given extra safeguards, including reduced exposure to graphic content, extreme
01:36beauty ideals, and an early warning system for self-harm signs.
01:41If triggered, a trained human team reviews the case and may alert parents or law enforcement.
01:45OpenAI acknowledges the system's imperfections, but believes it's better to send awkward notifications
01:51than remain silent.
01:53The AI may not always be perfect, but it's focused on caring when it matters most.
01:58Keyword T
02:03Expo
02:09Read
02:10Learn
02:12Read
02:13Eat
02:13You
02:13Mean
02:14噹
02:14judgment
02:14Read
Be the first to comment
Add your comment

Recommended