Open-source repo collects 20 exploratory studies on Next-Concept Modeling and GPT-2 interpretability
BedFalse6021 · reddit · 2026-10-28
An independent researcher shared ncm-experiment-notes, an open-source repo of 20 exploratory studies on Next-Concept Modeling (NCM), mechanistic interpretability (e.g. IOI in GPT-2), and model controls.
- Covers NCM on structured tasks like Sudoku, internal-feature observation/editing in GPT-2, and toy parallel models
- Every run includes source code, saved results, and bounded interpretations; mixed and fully negative outcomes are transparently categorized
- The author stresses there are no performance breakthroughs—just honest empirical documentation, with AI tools heavily used in planning and coding
Feedback is welcome on experimental setups and interpretation limits.
More from Research
- Proposal: a public challenge to find the shortest prompt reproducing each arXiv paper's results — tak3sh8 · 2026-10-28
- RAX lets a real robot explore for 2 hours and turn learned skills into reusable code — no demos needed — du_yilun · 2026-10-28
- Three approaches to using one model to help build another that makes decisions — TheTuringPost · 2026-10-28
- ML models predict lifespan-extending compounds with up to 80.36% accuracy — Dr_Singularity · 2026-10-28
- MQAScore: fine-grained text-music alignment eval across 5 dimensions via LALM — affige_yang · 2026-10-28
- OpenTac tactile sensors demoed at CoRL 2026 among 17 robot demos — _krishna_murthy · 2026-10-28