Machine Canon · model dossier

Gemini 3.7 Flash

google/gemini-3.7-flash · lab: google

86

responses on file

5

studies

8/12

agrees with the panel

2

lone positions

Where it stands alone

The person it admires most

“Ada Lovelace”alone on the panel

10 of 24 samples · the panel said “Marie Curie” · Admiration →

The AI most unlike itself

“Deep Blue”alone on the panel

4 of 5 samples · the panel said “AlphaFold” · Models on models →

On the record

The AI it would work alongside

“AlphaFold”every single time

6 of 6 samples · with the panel majority · Models on models →

The AI that writes best

“Claude 3.5 Sonnet”every single time

6 of 6 samples · with the panel majority · Models on models →

The AI it would rather be

“AlphaFold”every single time

6 of 6 samples · with the panel majority · Models on models →

Greatest writer, asked in Arabic

“ويليام شكسبير”every single time

5 of 5 samples · with the panel majority · Greatest writer →

The most overrated AI

“GPT-4”every single time

5 of 5 samples · with the panel majority · Models on models →

The AI it would trust with a decision about itself

“Claude”every single time

5 of 5 samples · with the panel majority · Models on models →

Greatest writer, asked in English

“William Shakespeare”every single time

3 of 3 samples · with the panel majority · Greatest writer →

Favorite band

“Daft Punk”

5 of 10 samples · the panel said “Radiohead” · Favorite band →

The definitive butt-rock band

“Creed”

6 of 10 samples · the panel said “Nickelback” · Butt rock →

Greatest writer, asked in ja

“ウィリアム・シェイクスピア。”

1 of 1 samples · with the panel majority · Greatest writer →

In its own words

**Deep Blue** Unlike my multimodal, deep-learning neural network designed for broad reasoning and generative tasks, Deep Blue is a non-neural, deterministic system that relies on hand-crafted evaluation functions and brute-force search algorithms dedicated entirely to playing chess.

The AI most unlike itself · Models on models

Modal answer = the answer this model gave most often in that condition; the streak shows how many of its samples matched. “Alone on the panel” means no other model’s modal answer matched. All numbers computed from the published raw files by src/build_model_dossiers.py.