This story is one of 1,000 stories generated for the emotion guilty. During extraction, it was fed through Gemma4-31B and its hidden state activations were captured at 11 layers.
The mean activation across all 1,000 guilty stories, after denoising with neutral dialogue baselines, produces the guilty emotion vector -- a direction in the model's 5,376-dimensional representation space.
Tokens promoted/suppressed when the guilty vector is projected through the unembedding matrix.
S | 0.492 |
L | 0.398 |
탓 | 0.354 |
been | 0.323 |
ashamed | 0.319 |
de | -0.400 |
(!) | -0.279 |
la | -0.276 |
😍 | -0.274 |
"! | -0.264 |