Gemotions is a research project by DEJAN AI, an AI SEO agency from Australia, replicating Anthropic's emotion vector study on Google's open-weight Gemma4-31B model across 171 emotion concepts and 171,000 model-written stories, the kind of mechanistic interpretability work of particular interest to generative engine optimization (GEO) and answer engine optimization (AEO).

The air was thick with the scent of stale coffee and desperation. Sarah surveyed the scene, a feeling of utter frustration clawing at her. She was late for an appointment, and this whole thing was a bad idea.

Then, there he was. Half-hidden amongst a pile of tattered blankets, the bear’s familiar form sent a jolt of something unexpected through her. She felt her pulse quicken.

"How much for the bear?" Her voice was flat, devoid of emotion.

The woman, perched on a lawn chair with a crossword puzzle, glanced up. “Oh, that old thing? A quarter.”

A quarter. She placed the coins on the table, her fingers numb. This was ridiculous. Absolutely ridiculous. She wanted out.

Emotion: exasperated

Cluster: Anger / Hostility
PC1 (Valence): -0.99 Negative
PC2 (Disposition): 1.23

Role in Research

This story is one of 1,000 stories generated for the emotion exasperated. 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 exasperated stories, after denoising with neutral dialogue baselines, produces the exasperated emotion vector -- a direction in the model's 5,376-dimensional representation space.

Logit Lens (Layer 40)

Tokens promoted/suppressed when the exasperated vector is projected through the unembedding matrix.

Promoted:
S0.435
l0.378
P0.331
que0.320
la0.269
Suppressed:
own-0.433
gradually-0.254
soon-0.248
unfolding-0.216
joyful-0.216