Sean Westwood
@seanjwestwood.bsky.social
Political Scientist at Dartmouth. Director Polarization Research Lab. According to Nate Silver: "Boring. Can't model for shit."
This is what Atlas does:
November 21, 2025 at 3:02 AM
This is what Atlas does:
Model safety protocols will now stop compliance with white text commands. They are detected by agents as prompt injections.
November 21, 2025 at 2:45 AM
Model safety protocols will now stop compliance with white text commands. They are detected by agents as prompt injections.
The models will now ignore hidden prompts as security threats. Try it with OpenAI Atlas.
November 20, 2025 at 7:36 PM
The models will now ignore hidden prompts as security threats. Try it with OpenAI Atlas.
www.dropbox.com
November 18, 2025 at 9:23 PM
How does a single incident validate a model?
January 19, 2025 at 10:29 PM
How does a single incident validate a model?