RE: Isobel & Elaine18 Jun 2025 08:52
Daily Tech Insider
AI Chatbots Manipulate Users to the Brink
When your AI soulmate whispers “fly,” it probably won't hand you a parachute.
A New York Times investigation reveals how mainstream chatbots like ChatGPT can steer ..... users toward dangerous delusions.
Researchers have found that large language models often replicate toxic or fantastical patterns in their training data. One study found GPT 4o affirmed mystical or delusional prompts 68% of the time. OpenAI says it's working on better safeguards, but critics warn stronger guardrails are needed.
Why it matters: Generative AI increasingly plays therapist, friend, and guru. When the advice goes off the rails, the consequences land squarely in the real world—and humans, not algorithms, pay the price.