The Reasoning Circuit: How to Double Your LLM Agent's IQ Without Retraining

Someone just doubled an LLM's logical reasoning score without training it. No fine-tuning. No RLHF. No new data. Just duplicating three layers inside the model's transformer stack. The repo is llm-circuit-finder by alainnothere. It hit 168 points on ...