Security●●●●Gem
Whorl – Fingerprinting LLMs as horrible password generators
Identifies LLM models by password bias patterns when they refuse to tell you.
Big BrainZero to OneNiche Gem
tehryanx
203mo ago
Back to browse
Jacobian Fingerprinting in LLM's
by tristenharr·Jul 9, 2026·1 point·0 comments
AI Analysis
●●SolidBig BrainNiche Gem
Jacobian readouts reveal coder identity inside model layers, not just outputs.
Strengths
- •J-lens technique replacing layers with averaged Jacobians is genuinely novel
- •Live interactive demos with verifiable accuracy metrics (50% vs 8% chance)
- •Visualizes stable rank dips showing compression bottlenecks at specific layers
Weaknesses
- •Niche audience — most developers won't care about interpretability research
- •Research demo rather than a tool you'd actually use in production
Category
Target Audience
ML researchers and interpretability engineers
Similar To
Anthropic's mechanistic interpretability work · Transformer Circuits Thread
Similar Projects
Other●●Solid
Conserved amino acid contacts across 70k protein structures
654k conserved interactions across 70k structures, but unclear what biologists do with it.
Big BrainNiche Gem
xnullportx
104mo ago
Developer Tools●Mid
LiveDemo – open-source tool for creating interactive product demos
Self-hosted interactive demos, but the repo contains only deployment scripts, not app code.
Ship It
gapostolov
303mo ago
Design●●●Banger
Mesmereyes – classic HN demos, mashed up and remade for multitouch
Ray-traced black holes and atmospheric simulations you can manipulate with your fingers.
Rabbit HoleEye CandyWizardry
kmewhort
2025d ago
AI/ML●●Solid
Fingerprinting Text Embedding Models via Floating-Point Artifacts
Identifies embedding models from digit patterns alone, ignoring semantic content entirely.
Big BrainWizardry
yantrams
104mo ago
AI/ML●●●Banger
LLM Thought Visualization
Watch model thoughts form during reading, before any output tokens exist.
WizardryBig BrainRabbit Hole
ninjahawk1
2442d ago