Open repo collects 20 next-concept modeling studies

An independent researcher open-sourced 20 exploratory studies on Next-Concept Modeling, GPT-2 mechanistic interpretability and model controls, with full code and results including negative findings. The project claims no performance breakthroughs and stresses reproducibility and controls.

• Building GPT-2 KV Cache from Scratch and Measuring Performance2026-10-18
• Redditor revisits 2019 GPT-2 'too dangerous to release' headline2026-10-13
• Experimenting with hypersurface-constrained dynamic weight updating2026-09-19
• Discussion: GPT-2 Not Fully Released Over Misuse Concerns2026-09-14