Verify LeCun's account of DjVu's free-books purpose and the free NIPS 0–13 archive, and check whether nips.djvu.org has rotted

djvulecunnipsfree-knowledgelink-rotverificationhistory-of-mlinternet-archive

This capture answers the routed question at question-verify-lecun-djvu-free-nips-archive, which flagged claim-djvu-created-to-freely-distribute-scanned-documents [unverified-quote] because its only source — a LeCun X/Twitter post — had returned HTTP 402 at capture time and reached the vault only as a search-engine summary, with no verbatim phrasing preserved. This session re-fetched the post directly with archive_page (it resolved cleanly this time) and cross-checked the archive's own historical content and current reachability against the Wayback Machine and the official NeurIPS proceedings site.

Claim: LeCun's own account (in his own words, on his own account) states DjVu's purpose was cheap high-resolution distribution of scanned documents over the "newly expanding" internet, demonstrated by freely releasing the NIPS proceedings

Claim type: historical / definitional (what LeCun says his own project was for). Source tier required: Tier 3–4 acceptable for an uncontested historical claim, but since this is the founder's own testimony about his own project, Tier 1 is available and used.

In a post on X dated 2 January 2024, Yann LeCun states directly: "In the mid 1990s, I started a project called DjVu at AT&T Labs. The purpose was to devise a new image compression format so that printed documents could be scanned at high resolution and distributed efficiently over the newly expanding Internet." He frames this explicitly as a story "about free books." The account is now fetched and read directly (not relayed via search summary), and the verbatim quote is preserved above with source_sha — this resolves the [unverified-quote] flag on claim-djvu-created-to-freely-distribute-scanned-documents for this portion of the claim.

One correction surfaces against the existing flagged note: it records source_date: 2023-01-03. The post itself is timestamped "7:41 PM · Jan 2, 2024" (with the reply thread continuing into Jan 3, 2024) — a full year off. The date should read 2024-01-02, not 2023-01-03.

Provenance: Yann LeCun (@ylecun), post on X, 2 January 2024. https://x.com/ylecun/status/1742269871168111018

Claim: LeCun states the NIPS proceedings were scanned with publisher permission and posted free at nips.djvu.org starting in 2000, as a demonstration of DjVu

Claim type: historical, with an embedded quantitative element ("13 volumes"). Source tier required: Tier 1–2 for the number.

LeCun's own account: "As a useful demonstration of the technology, I decided to scan and distribute the complete collection of proceedings of the Neural Information Processing conference (NIPS). I asked the publishers, Morgan Kaufman and MIT Press, for permission to do that. They agreed because they weren't making any revenue from past proceedings. We scanned the 13 volumes, OCRed and indexed all of the material, and put it up on a free website in 2000: nips.djvu.org." A reply in the same thread adds that Volume 0 (NIPS 1987) had been published by the nonprofit American Institute of Physics, whose rights had since passed to Springer, and that LeCun's team separately sought Springer's permission for that volume.

This is a Tier 1 source (the founder's own account) for a quantitative figure ("13 volumes"), clearing the sourcing floor. There is a minor, unresolved discrepancy worth flagging rather than silently resolving: LeCun's tweet does not itself state which 13 volumes, but a 2024 crawl of the live site's own table-of-contents page (see next claim) lists 14 numbered volumes — 0 through 13, spanning NIPS 1987 through NIPS 2000 — not 13. It is plausible the original 2000 launch covered 13 volumes (0–12, i.e. 1987–1999) and Volume 13 (NIPS 2000, published by MIT Press in 2001) was added afterward, but this session found no source that states the count explicitly at both points in time, so the discrepancy is recorded, not resolved.

Provenance: Yann LeCun (@ylecun), post + reply thread on X, 2–3 January 2024. https://x.com/ylecun/status/1742269871168111018

Claim: nips.djvu.org no longer resolves as a live website (checked 2026-08-13), though the Wayback Machine preserved its content through at least November 2024

Claim type: technical-mechanism / historical (a reachability claim about a specific domain, checked directly). Source tier required: Tier 1–2 — met by direct, repeated fetch attempts against the live domain.

Two independent fetch tools were used to test nips.djvu.org directly on 2026-08-13: mcp__seek__archive_page failed with urlopen error [Errno 8] nodename nor servname provided, or not known (a DNS resolution failure) on three separate attempts (root domain over both HTTP and HTTPS, and a specific /nips-toc.html path), and WebFetch independently failed with getaddrinfo ENOTFOUND nips.djvu.org on the same path. Both are DNS-resolution failures, not server errors — the domain name itself does not currently resolve, which is the strongest available signal (short of a manual whois lookup, not performed) that the domain has lapsed or been abandoned outright, not merely that its server is down.

This is not total loss: the Internet Archive's CDX index shows Archive Team crawled and preserved pages from nips.djvu.org — including per-volume tables of contents — as recently as 19 November 2024. A Wayback Machine copy of nips-toc.html (2024-08-17 capture) confirms the site's own content: a volume selector listing Volume 0 (NIPS 1987) through Volume 13 (NIPS 2000), each linking to a full table of contents. The demo Yann LeCun describes did exist and was crawled by the Archive as recently as late 2024; what has rotted is the live domain itself, not (yet) all record of what it once contained.

Provenance: direct fetch attempts against https://nips.djvu.org/ and http://nips.djvu.org/nips-toc.html, 2026-08-13 (this session); Wayback Machine CDX index at http://web.archive.org/cdx/search/cdx?url=nips.djvu.org*; archived snapshot at http://web.archive.org/web/20240817135443/http://nips.djvu.org/nips-toc.html

Claim: The content of the free NIPS archive was not simply lost when the domain died — it was substantially absorbed into the official NeurIPS proceedings site, which now hosts the same early volumes (0 onward) as its own permanent record

Claim type: historical, corroborating LeCun's own account of the archive's eventual fate. Source tier required: Tier 3–4 acceptable (uncontested institutional fact); Tier 1 available and used.

LeCun's account states: "Eventually, the NIPS conference stopped making printed proceedings and started hosting all the books on their website (nips.cc), including our scans." This is independently corroborated: the official NeurIPS proceedings index at papers.nips.cc, fetched directly in this session, lists a complete run from "Neural Information Processing Systems 0 NIPS 1987" and "Advances in Neural Information Processing Systems 1 NIPS 1988" forward through NeurIPS 2025 — meaning the same early-volume content the DjVu demo pioneered making freely accessible is now permanently hosted at the conference's own institutional domain, independent of nips.djvu.org's survival. The free-access outcome LeCun describes as the point of the demo persisted structurally even though the original demonstration site did not.

Provenance: papers.nips.cc, "List of Proceedings," fetched 2026-08-13. https://papers.nips.cc/

Answer to the core question

Confirmed, with one correction and one open discrepancy. LeCun's account of DjVu's free-books purpose and the free NIPS 0–13 (effectively 0–12 or 0–13, see discrepancy above) archive is now verified against his own words, directly fetched and hashed — the [unverified-quote] flag on claim-djvu-created-to-freely-distribute-scanned-documents should lift. The source_date on that note (2023-01-03) is wrong and should be corrected to 2024-01-02. nips.djvu.org has rotted: it fails DNS resolution as of this session (2026-08-13), confirmed via two independent tools — the sharper finding the original question anticipated. But it has rotted into the Wayback Machine's care (crawled as recently as November 2024) and, more importantly, its actual mission — free access to the early NIPS proceedings — now lives on at the conference's own official site, papers.nips.cc, independent of the original demo domain's survival.

Further leads

• LeCun's reply-thread side story about NIPS Volume 0 (1987) — originally published by the nonprofit American Institute of Physics, rights later sold to Springer, requiring a separate permission request — was only partially readable (the reply was truncated with "Show more" in the archived capture); worth a direct fetch of the reply tweet itself for the full text.
• The DjVu format's technical mechanism (IW44 wavelet codec for continuous-tone layers, JB2 for the bitonal text mask) is already flagged unverified-against-primary in claim-djvu-1998-paper-coauthored-by-lecun-and-bengio; the 1998 primary paper (available at yann.lecun.com/exdb/publis/pdf/lecun-98c.pdf) was not fetched this session and remains the way to close that gap.
• Whether a whois lookup on nips.djvu.org would show the domain registration itself lapsed (vs. a DNS/hosting misconfiguration on an otherwise-live registration) was not checked this session — would sharpen "rotted" from "unreachable" to "abandoned."
• LeCun's claim that the ML community's "culture of fast posting of preprints and open-access publications" traces partly to this NIPS/DjVu episode and to JMLR's founding as an early open-access journal is a broader historiographic claim, asserted but not independently corroborated this session.

Entity candidates

• Morgan Kaufmann Publishers — concept/organization — one of the two publishers LeCun says granted permission to scan and freely release the NIPS proceedings; the rights-holder whose forbearance the whole "free books" demo depended on.
• MIT Press — concept/organization — the other publisher granting permission per LeCun's account, and (per the archived table of contents) publisher of NIPS Volume 13 (2000) directly; also the current publisher of record for later NeurIPS proceedings.
• American Institute of Physics (AIP) — concept/organization — original nonprofit publisher of NIPS Volume 0 (1987), whose rights had since passed to Springer per LeCun's reply thread; the earliest rights-holder in the chain the free-archive project had to negotiate around.
• Springer — concept/organization — for-profit successor rights-holder to AIP for NIPS Volume 0, separately contacted per LeCun's account; a useful contrast case (for-profit vs. nonprofit permission-granting) within the same story.
• Léon Bottou — person — DjVu co-author (already the connective figure in claim-djvu-1998-paper-coauthored-by-lecun-and-bengio) not yet an entity page in this vault; worth flagging given how often he recurs across the DjVu and Perceptrons clusters.
• Internet Archive / Archive Team — concept/organization — the entity that crawled and preserved nips.djvu.org's content after the fact; already connected via claim-internet-archive-2016-stopped-generating-djvu, now also relevant as the archive-of-the-archive for this specific domain's rot.