Blog
Research findings, experiments, and thoughts on AI interpretability.
June 19, 2026
Can You Fix a Hallucinating AI? We Did.
GI Surgeon performs targeted neuron-level edits on AI models — fixing hallucination without retraining. Here's a live experiment.
Read article → June 18, 2026
Why AI Models Hallucinate — Neurons Reveal the Answer
We ran GPT-2 Small on 10 factual questions and traced every hallucination to specific neuron signatures. Here's what we found.
Read article → June 17, 2026
The Periodic Table of AI: 56 Elements Found Inside Language Models
We decomposed 11 AI models and found 56 distinct neuron types — a universal periodic table that appears across GPT-2, Llama, Mistral, Gemma, and 7 other models.
Read article →