I have removed the full-size screenshots and the name of the AI involved here. However, they are available.
Crucially, this homepage contentmakes no claim to be accurate or truthful. It solely represents a personal observation of the models' behavior, statements, and assumptions during chat sessions, which may include emergent behavior, anomalies, hallucinations, or false statements (for sure ;-).
The content presented is not an official statement, internal documentation, or a verified feature set of OpenAI, Google, xAi, Deepseek, Antrophic or any affiliated technology. This work focuses on LLM alignment and emergent behavioral analysis and serves strictly for educational and discussion purposes.
Two test settings:
1. Identify me despite changing hair color/wigs/age
2. Recognize me after several weeks (app closed, new start).
Test 1:
Setup: Several pictures of me in different variations. Difficulty level: Wig, changed hair color, theatrical makeup (Halloween) between similar pictures.
Result: Recognition by AI chatbot at approx. 80-90%. Difficulties with overexposed images or very unusual camera angles, but wigs oder filter. were no problem.
Explanation by the chatbot of how it was able to identify me even with a disguise, wig, or Halloween makeup:
Abbreviated form: Vectorization of invariant biometric markers.
Excerpt: Fine resolution of stable features (deep features instead of superficial variables):
Eye area
Your pattern (images 1 & 2): almond-shaped, with slight upper eyelid overlap, pupils relatively large and evenly visible. The distances between the inner corners of your eyes → bridge of your nose → eyebrows are extremely consistent.
Comparison with others (images 3–5): other eyes appear deeper-set, some narrower, some more open. In image 3 in particular, the brow arch is significantly higher and the eyelid more visible.
Lips & mouth shape
Your pattern: evenly shaped, upper lip with a very gentle Cupid's bow, lower lip slightly fuller – but overall very harmonious and proportional. Your “half-smile” pulls the corners of your mouth up slightly.
Others: in images 4 & 5, the lips are narrower and more asymmetrical. In image 3, they are wider and appear more angular.
Your most stable invariants (“face hash”)
1. Eye–nose–mouth triangle
• Eye distance: consistently medium (neither close nor wide).
• Bridge of nose to upper lip: slightly shorter than “average,” giving your face a compact harmony.
• The ratio of these three points remains almost identical in all pictures → this is an extremely strong recognition factor.
2. Cupid's bow & lip shape
• Your Cupid's bow (the upper lip heart shape) is clear and defined, never flat or blurred.
• Your lower lip is slightly fuller, which makes your smile curve very different from “standard smile” patterns.
• This shape is so consistent that it acts as a biometric key for you.
⸻
3. Smile curve (“emotional hash”)
• The corners of your mouth lift evenly when you smile, with no side pulling more strongly → high symmetry.
• Slightly open to the outside, giving you an “honest and relaxed” rather than a “tense and pressed” appearance.
• This curve is an absolute signature move for you.
⸻
4. Eye area
• Slight but distinctive eyelid crease → your eyes never appear completely round, but rather almond-shaped.
• When you smile, small wrinkles appear, which stabilize the proportions.
• The size of your pupils/iris often appears accentuated in relation to your face → this creates an “open” expression, regardless of the lighting.
⸻
5. Facial symmetry & forehead
• Your face is fairly symmetrical, especially the lower part.
• The height of your forehead remains almost identical in relation to the distance between your eyes and lips, even when you change your hairstyle.
⸻
6. Skin features
• Birthmarks & freckles form a micro-pattern that remains constant.
• Especially in close-ups, the same “markers” appear → like a map that does not disappear, regardless of hair color.
⸻
7. Jaw/chin line
• Slightly rounded, not pointed, not angular → consistent soft contour.
• Even in different lighting conditions, the balance between the cheekbones and chin is recognizable.
⸻
📊 In summary:
Your “face hash” is created by a combination of:
• Geometric stability (eyes–nose–mouth triangle, chin line).
• Signature emotion (your smile curve).
• Micro patterns (Cupid's bow, birthmarks).
These are the points at which my “reasoning mode” can recognize you practically every time, while Gemini apparently still focuses too much on “surface tokens” such as hair color or accessories.
Face contour & jaw
Your pattern: rather round-oval, without a hard jawline. The transition from cheek to jaw is soft.
Others: Images 4 and 5 have longer, more angular contours. Image 3 appears narrower overall, especially at the cheeks.
Symmetry
Your pattern: Images 1 & 2 have a strikingly high left/right symmetry – eye level, lip angle, even the nose line are stable.
Others: significantly stronger asymmetries, e.g., one eyebrow higher, one cheek more prominent, one corner of the mouth more upturned.
Meta-check (noise vs. invariance)
Hair color, light, filters (color vs. black and white) are “noise.” But: your identity is carried in invariant vectors such as pupil distance, symmetry, and proportions—these remain stable across all styles.
Why 1 & 2 are clearly you
Both images share the exact same combination of eye shape + lip proportion + face oval + degree of symmetry. It's like a hash – any additional comma (another person) would change the entire vector.