Tagged "AI safety"
The Bicameral Model
May 11, 2026   |   17 min read   |   technical
Two frozen language models wired together through a small trainable interface on their hidden states, so they can share their thinking as it happens instead of trading text.