Steering LLM activations along a learned manifold (e.g., the affective circumplex) is significantly more effective than off-manifold steering.

AI models and biological brains represent concepts with similar geometric structure, as demonstrated by measurable alignments between animal brain neural recordings and AI model internals.

“If you look at how do how do these AI models represent just like represent concepts and you look at the parts of the brain that represent concepts, you see these you see very similar geometry. That to me was one of the first clues. Like when I really saw that and we use that practically. Like we use that constructively at science. Like that's we know that is that is true because we can get alignments between animal brain neural recordings and AI model internal.”

Following the low-dimensional curve of an emotion manifold always yields consistently emotionally adjusted responses for any input.

“If you follow the curve, you follow the wheel, you can always consistently get like emotionally adjusted responses for basically any input.”