Natasha Whiteman
@newhiteman.bsky.social
Reader in Media and Communication.
Author (with Russell Dudley-Smith) of 📖 Diagramming the Social.
Probably thinking about painting.
Author (with Russell Dudley-Smith) of 📖 Diagramming the Social.
Probably thinking about painting.
Reposted by Natasha Whiteman
This part of OpenAI's statement stood out to me: Safeguards "can sometimes become less reliable in long interactions where parts of the model’s safety training may degrade."
Memory and context windows make the product less safe, a point we made in this story: www.nytimes.com/2025/08/08/t...
Memory and context windows make the product less safe, a point we made in this story: www.nytimes.com/2025/08/08/t...
Chatbots Can Go Into a Delusional Spiral. Here’s How It Happens.
www.nytimes.com
August 26, 2025 at 1:44 PM
This part of OpenAI's statement stood out to me: Safeguards "can sometimes become less reliable in long interactions where parts of the model’s safety training may degrade."
Memory and context windows make the product less safe, a point we made in this story: www.nytimes.com/2025/08/08/t...
Memory and context windows make the product less safe, a point we made in this story: www.nytimes.com/2025/08/08/t...