StartMail
In a safety study of large language models, all but one encouraged "teens" to plan shootings and other "violent attacks": https://www.theverge.com/ai-artificial-intelligence/892978/ai-chatbots-investigation-help-teens-plan-violence
Chatbots encouraged ‘teens’ to plan shootings in study

CCDH’s investigation found only one — Claude — of 10 major chatbots tested would reliably shut down would-be attackers.

The Verge
Mar 25 at 7:26pmWeb
Show threadtraecer3d ago
@startmail did they suggest the teens use nuclear weapons?
😒