Skip to content

Blog Posts

Use this file for public explainers and blog-style summaries.

Assume This Reader

  • Assume general familiarity with LLMs, training, tokens, and transformers.
  • Do not assume deep knowledge of the specific technique.

Optimize For

  • Explain what the work is and why it matters.
  • Preserve technical accuracy without turning the post into a paper.
  • Make the result legible to a strong ML engineer outside the niche.

Write This Way

  • Explain new or non-standard concepts before leaning on them. Linking to explanations elsewhere is acceptable.
  • Favor intuition, framing, and observed results over derivations.
  • Keep implementation details light unless they are central to the point.
  • Use examples and concrete comparisons instead of abstract claims.
  • Lead paragraphs with the main takeaway, not setup.
  • Prefer concrete numbers or comparisons over qualitative claims.

Avoid These Failure Modes

  • Do not drown the reader in internal terminology.
  • Do not hide the result behind long context-setting.
  • Do not turn the post into marketing copy.
  • Do not imply broader conclusions than the data supports.

Test

  • Ask whether a strong ML engineer outside the subarea could follow the piece without looking things up.