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 clinking of champagne flutes carried from the opulent garden party next door, a stark contrast to the usual lawnmower soundtrack. I glanced through my kitchen window. There was Mr. Finch, the quiet accountant, wearing what appeared to be a tuxedo, surrounded by people in expensive dresses. A chuckle escaped my lips.

I decided my evening needed a twist. I donned my oldest, most patched-up gardening clothes, grabbed a watering can, and marched over. I pretended to be incredibly engrossed in my neighbor's prize-winning petunias.

"Oh, these are lovely!" I exclaimed, my voice a little too loud. "Do you have any tips? I just can't seem to get mine toโ€ฆ thrive." I knew my eyes were sparkling.

Emotion: playful

Cluster: Positive / Joy
PC1 (Valence): 4.42 Positive
PC2 (Disposition): 1.44

Role in Research

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

Logit Lens (Layer 40)

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

Promoted:
B0.662
(!)0.566
!0.562
!?0.553
๐Ÿ˜œ0.553
Suppressed:
ness-0.685
๐Ÿ˜ž-0.590
S-0.562
๐Ÿ˜”-0.470
ะธะดั‚ะธ-0.461