Open Postprint

Karpathy's long X post on working with language models

Updated 2026-10-13 · every figure below is a real sample taken on this date

Today's X long-post sample has an unusual pairing at the top: an individual researcher and a lab's own announcement, both past a thousand characters. The researcher's post won on likes; the lab's won on reach.

What the sample collected

The Postprint daily run took a batch on 2026-10-13 (Asia/Shanghai) from twenty public X accounts, including @pmarca, @balajis, @karpathy, @paulg, @sama and @naval, using the public read-only FxTwitter API. It observed 109 items, verified and published 4, and excluded 2 on editorial grounds. Entries are ordered by a composite score — likes, plus twice the reposts, plus three times the bookmarks, plus a small views term — not by any single counter. All interaction numbers are real sampled values with their collection time.

The post at number one

Andrej Karpathy (@karpathy) posted a long-form entry of about 1,816 characters opening with the line that there will be more time spent trying to understand the outputs of language models. Sampled metrics:

  • 55,313 likes · 7,795,707 views
  • 6,512 reposts · 1,621 replies · 1,496 quotes
  • 81,444 bookmarks

The bookmark figure is the striking one. Bookmarks outnumber likes by roughly one and a half to one — the signature of a post people intend to use rather than merely endorse. Practical advice gets bookmarked; opinions get liked. The post, as sampled, is a set of working notes on writing and prompting, including the suggestion to ask a model to explain something in a controlled language specification.

The runner-up, and why the two are not comparable

Second place went to @AnthropicAI, announcing a threat-intelligence report covering attempted misuse of its model. Its counters are a different shape entirely:

  • 50,462 likes · 43,803,534 views
  • 11,706 reposts · 3,227 replies · 4,935 quotes

Roughly similar likes, more than five times the reach. An institutional announcement is surfaced by the platform's distribution; a practitioner's notes spread through people deciding they are worth keeping. Both are interesting, but reading them as a single ranking would flatten exactly the difference that matters.

Why long X posts are worth keeping

Long posts are the part of X most likely to disappear from your view. They scroll past, the timeline truncates them, and the useful ones — the ones people bookmark — are precisely the ones you want later. Keeping them as documents rather than links also protects you from link rot on work you may cite.

How to export one

1Copy the post linkShare menu on mobile, address bar on desktop.
2Paste it into PostprintPublic posts need no login.
3Choose PDF, Word or MarkdownMarkdown is the better choice if the text is going into notes or another tool.
4Keep the source URLThe export carries it, so the citation stays intact.
Bookmarks are a good proxy for “worth saving” when you are deciding what to archive. A post with more bookmarks than likes is usually a how-to.

What this page is not

  • Metrics are point-in-time samples, not live counters.
  • The headline above is our summary of the post's opening, not the author's own title.
  • This is a twenty-account observation, not an X-wide ranking.

Postprint

Paste a public X post link and export it to PDF, Word or Markdown with its source link intact.

Open Postprint

More from Postprint

How to save an X (Twitter) Article as a PDFX Articles & Long Posts Daily Picks — 2026-10-13How to save a Twitter (X) thread as a PDFPostprint — export an X post to Word, PDF or Markdown