🚀 Latest Interactive Research
Frequency Domain Neural Networks: An Exploration
Could neural networks operate fundamentally in frequency domain? We built three interactive demos exploring the activation function barrier, phase dominance in images, and the mysterious 90% sparsity cliff.
Featured Tool
Prompt Diff
Compare two text outputs with visual diff highlighting. A simple tool for evaluating prompt variations and LLM output changes.
🎮 Interactive Research Demos
Explore cutting-edge research through hands-on visualizations. Click, adjust, and discover insights.
Activation X-Ray Demo
InteractiveWatch how activation functions destroy frequency sparsity in real-time. See why ReLU is the enemy of frequency domain efficiency.
Phase vs Magnitude Showdown
InteractiveDiscover why phase dominates image structure. Swap frequency components between images and see the surprising results.
Sparsity Cliff Demo
InteractiveExperience the counter-intuitive reality: 90%+ sparsity required for real speedups. Quality drops smoothly, speed jumps suddenly.
Recent Work
Frequency Domain Neural Networks: An Exploration
A deep dive into frequency domain operations, activation function barriers, and three interactive demos that reveal surprising insights.
Lessons from AI Consciousness Detection Research
We attempted to build measurable tools for detecting consciousness in AI systems. The approach had fundamental flaws, but the process revealed valuable insights.
Building Research-Critic: A Tool Development Story
Creating external AI validation for research methodologies - a tool that proved its worth by catching flaws in our own validation claims.
Fast ablation harness for prompts
A CLI script to test prompt variations in parallel, eliminating hours of manual repetition.
Working with uncertainty
Most AI work lives under uncertainty. Treat it like a first-class constraint: state assumptions, bound risk, and iterate in tight loops.
Against hand-wavy AI evaluations
Most AI evaluations are theater. Here's how to build evaluations that actually measure what matters.
Mission
📝 Research in Public
Document experiments, failed approaches, and incremental progress. Make the research process transparent.
🔬 Question Hype
Test claims rather than repeat them. Build tools to measure what matters, not what sounds impressive.
🛠 Ship Useful Tools
Create micro-tools that solve real problems for practitioners. Open source everything.
📊 Share Complete Results
Publish both positive and negative findings. Document what works, what doesn't, and the methodology behind each conclusion.
Current Focus
- Exploring measurable AI capabilities and behavioral research methods
- Iterating on Prompt Diff tool based on user feedback
- Weekly lab notes documenting experiments and learnings
Stay Updated
Follow via RSS for updates on research, tools, and experiments. All work is documented transparently with full methodology.