30papers.com – Ilya's 30 essential ML papers, in a beginner friendly format

Original Article ↗ Hacker News Discussion ↗

Purpose and Origin of the Site

  • Built by a first-year CS student as a side project to help friends get into reading ML papers.
  • Initially intended as a small, informal resource; surprised by the volume of attention.
  • Current site mostly aggregates the list and hosts reformatted versions of papers; intent to expand is still evolving.

Source and Legitimacy of the Paper List

  • List described as based on a “rumored” set of papers given by a well-known researcher to another well-known engineer.
  • Some commenters question the authenticity, noting multiple circulating lists, a lack of clear sourcing, and that the original story mentioned ~40 papers.
  • Others point to posts on X and a commercial “Sutskever’s List” book as related or reinforcing.
  • Several conclude that, regardless of provenance, the selected papers are widely regarded as high quality and pedagogical.

UX, Design, and Accessibility Feedback

  • Strong criticism of the main page’s heavy animations, background motion, and scroll behavior; some report dizziness and headaches.
  • Suggestions: respect reduced-motion preferences, stagger animations, simplify or remove effects, enlarge fonts, reduce header height, and provide a plain list view.
  • The author responds by adding toggles to disable motion and complex backgrounds, but some still find fonts and navigation problematic.

Paper Rendering and “Beginner-Friendly” Claims

  • Complaints that LaTeX is poorly rendered, with subscripts/superscripts flattened and images/tables missing, making math hard to follow.
  • Several argue that in this state, it’s better to just link to arXiv rather than partially re-render PDFs.
  • The “beginner friendly format” label is questioned: the content is still math-heavy and not obviously scaffolded.

Requests for Annotations, Ordering, and Context

  • Multiple users expected short explanations or annotations summarizing the key ideas and what the author learned.
  • Requests for:
    • A clear statement of the site’s goal.
    • Chronological or “logical” reading order (e.g., early attention papers before Transformers; foundational ML before modern deep learning).
    • Proper citations (authors, year, venue).
  • The author is open to adding annotations and reflections, notes having informal ones elsewhere, and invites contributions while warning they are not formally trained in ML.

Alternative Resources and Related Work

  • Commenters share:
    • Plain link lists extracted from the site’s JSON.
    • Shell commands and Zotero tips for bulk downloading and organizing papers.
    • Previous blog series giving layman’s summaries of the same set of papers.
    • Other recommended intros (e.g., illustrated AI guides, classic courses, and explainers like CS231n, “Understanding LSTMs”, “Unreasonable Effectiveness of RNNs”).

Community and Meta Discussion

  • Some praise the site as a beautiful, motivating way to surface important ML work.
  • Others are highly critical, calling it clickbait, LLM-level curation, or “vibe-coded” at the expense of usability.
  • A substantial subthread debates Hacker News culture: harshness vs. constructive criticism, adherence to community guidelines, and the emotional impact on new creators.