FIG · License
Free to read. The work is ours.
obvix-learn is free to read in full, every chapter, no signup. The concepts belong to the field and we credit where we learned them. The writing, the figures, and the code are our own work.
FIG · Free
Read and learn
Every chapter, figure, and notebook is free to read, in full, with no account.
- Read all of it, as much as you like, at no cost.
- Quote a line or a figure for teaching or review, with attribution.
- Learn the ideas and write your own. The concepts are the field's, not ours.
FIG · Ours
© Obvix Labs
The writing, the figures, the code, and the way the 27 chapters are arranged are original work, © Obvix Labs.
- Please don't republish, mirror, or repackage that work wholesale.
- Please don't rebrand it or sell it as your own course.
- We'd rather you didn't train models on it. Ask us first.
How to cite a chapter
In a footnote or bibliography, the canonical form is:
Karan Prasad. "Ch. 22: Mechanistic Interpretability." obvix-learn, Obvix Labs, 2026. https://learn.obvix.io/chapters/mechanistic-interpretability
For a screenshot, slide, or talk, "Karan Prasad, obvix-learn (Obvix Labs, 2026)" is enough.
What we do and don't claim
We wrote the explanations, drew the figures, and wrote the code from scratch, so that expression is ours to license. The ideas underneath (the math, the algorithms, the techniques) belong to the field. We learned them from the sources we credit, and anyone else is free to learn them and write their own. All we ask is that you not lift our writing or figures wholesale. For translation, classroom use, or anything past quoting, write to connect@obvix.io.
Trademarks and identity
The obvix wordmark, logo, and visual identity are ours. Use them in a way that signals affiliation only when affiliation actually exists.
Third-party runtime assets
A few optional learning features execute open-source runtimes or models in your browser. Those works remain under their own licenses; our copyright statement does not replace or restrict them.
- DistilGPT2 (82M parameter model) — Hugging Face model, served as the pinned Xenova ONNX conversion in the optional browser demo. Apache License 2.0 ↗
- Transformers.js 3.7.1 — Hugging Face browser inference library, self-hosted for the optional GPT-2 demo. Apache License 2.0 ↗
- ONNX Runtime Web — WebAssembly inference runtime bundled by Transformers.js. MIT License and upstream third-party notices ↗
- Pyodide 0.26.4 — Browser Python runtime loaded for supported interactive labs. Mozilla Public License 2.0 ↗