Technical Blog.
All articlesSemantic
Pareidolia
Direct Image Optimization
A frozen neural network. A trainable image. Follow the pixels, embeddings, and gradients behind an image learning a new meaning.
Read the visual essayYouTube
Algorithm
Reverse Engineering the Recommender
Explore candidate retrieval, ranking objectives, content encoders, expert mixtures, and stochastic exploration in video recommendations.
Read the visual essayDream-Based
Learning
Synthetic Dreaming & Memory Consolidation
Can a model learn from the conversations it has? Explore a prototype that turns past experience into possible future dialogues, then uses them as training data.
Read the visual essayResearch
Explorations in AI architecture, latent space dynamics, and bio-inspired learning.
Implementing Semantic Pareidolia via Direct Image Optimization
Forcing neural networks to hallucinate abstract concepts into existing images using CLIP.
Latent Manipulator: Stochastic Torsion in High-Dimensional Latent Manifolds
Explores the application of non-Euclidean torque to pre-trained generative adversarial networks (GANs).
LLM Dream-Based Learning
A prototype that turns past conversations into possible future dialogues for training.
Reverse Engineering the YouTube Algorithm
Tracing the evolution of the recommendation engine from 2016 to today.
Trusted By
