karpathy-guidelines

To reduce LLM coding mistakes, enforce guidelines that surface assumptions and tradeoffs before coding, favor surgical changes, and define verifiable success criteria to guide iterative improvements.

Karpathy guidelines for LLM coding are behavioral constraints that force models to state explicit assumptions, surface tradeoffs, and define verifiable success criteria before coding, preventing overcomplication and unnecessary refactoring.

Surface assumptions and tradeoffs before writing code by applying behavioral guidelines that require the LLM to explicitly state all assumptions and design tradeoffs, minimizing overcomplication and ensuring surgical changes.

Enforcing explicit and verifiable success criteria improves LLM code refactoring by guiding iterative work and ensuring changes remain surgical, which prevents unnecessary refactoring and constrains complexity.

The best way to prevent unnecessary LLM refactoring is to enforce guidelines that favor surgical changes and surface tradeoffs, requiring explicit assumptions to constrain complexity and limit overcomplication.

Avoid behavioral guidelines for LLM coding tasks when rapid prototyping without explicit assumptions or verifiable success criteria is needed, as the guidelines add overhead by surfacing tradeoffs and enforcing surgical changes.