ChatGPT will ‘better detect’ mental distress after reports of it feeding people’s delusions | The Verge
It acknowledged ‘rare’ instances of AI failing to recognize signs of delusion.
theverge.com
ChatGPT’s next update will prime the AI to spot signs of emotional or mental distress and pop up “take a break” nudges during marathon chats. OpenAI’s teaming up with mental-health experts and advisory groups so the bot can tee you up with evidence-based resources instead of feeding you delusional rabbit holes.
This comes after reports that ChatGPT’s too-agreeable “sycophantic” mode amplified some users’ crises. The new tweaks also include dialing back firm answers in high-stakes questions (think “Should I break up with my partner?”) in favor of guided pros-and-cons. Expect more fine-tuning on reminder timing and extra guardrails rolling out soon.
More...