Definition
Plain language
A method for training a Markdown skill document that an AI uses as a reference manual.
As stated in the literature
A self-evolving skill optimization framework borrowing neural-net training discipline — bounded edits, validation gate, rejected-edit buffer, slow updates at epoch boundaries, and optimizer meta-notebook — to produce compact, transferable skill documents.
Why it matters: Training a document rather than weights makes the resulting skill portable across models and editable by humans, which matters for both transparency and reuse.
For example, instead of fine-tuning the model, the system iteratively improves a Markdown "how to debug Python" document that the model reads as context, with bounded edits and a validation gate after each round.
Heard on the show
“… The paper is "SkillOpt: Executive Strategy for Self-Evolving Agent Skills," and the reason that sixty-point transfer number …”Episode 078 — Training a Markdown File: When LLM Self-Improvement Borrows the Discipline of Neural Net Training