The Compiler's Ceiling¶
Abstract¶
For most of recorded history the great reference works were, surprisingly often, the labor of a single obsessive mind. Pliny, Isidore of Seville, Vincent of Beauvais, William Smellie, Samuel Johnson, Noah Webster — each set out, more or less alone, to fold a large slice of the world's knowledge into one coherent object. And each ran into the same wall: a ceiling of a few million words assembled over a few decades, because the binding constraint was never the availability of knowledge. It was a human being's finite lifetime of reading a source and setting it down again. That ceiling has just been lifted, and by roughly an order of magnitude. This essay is about what the lifting does and does not mean — and it argues that the newly reachable volume is the least interesting thing about it, because the scarce ingredient in an encyclopedia was never words. It was the single sensibility that made a million entries feel like one work.
The ceiling, and who hit it¶
The pattern is old and remarkably consistent. Pliny the Elder's Natural History (AD 77) — the work often called the first encyclopedia — runs to something on the order of four hundred thousand words across thirty-seven books.[1] It was compiled by one Roman official who, his nephew reports, kept a secretary reading aloud to him through nearly every waking hour, even at meals and while he was being rubbed down after a bath.[2] Isidore of Seville assembled the Etymologiae alone in the early seventh century and handed the West a reference shelf it leaned on for the better part of a millennium — a book that survives in more manuscripts than anything of its era but the Bible.[3] Vincent of Beauvais spent roughly twenty-nine years on the Speculum Maius and reached about four and a half million words — the largest reference work produced in the West before 1600, and very likely the high-water mark for what one compiler could physically accumulate in a lifetime.[4]
The lexicographers hit the same wall from the other side. Samuel Johnson wrote his Dictionary of the English Language (1755) over roughly nine years — forty-odd thousand headwords braided through with some hundred and fourteen thousand illustrative quotations he chose by hand, and very nearly single-handed.[5] Noah Webster gave his American Dictionary about twenty-eight years and seventy thousand entries, learning twenty-six languages along the way to get the etymologies right.[6] Even the Encyclopædia Britannica began, in 1768, as very nearly a one-man effort: William Smellie compiled almost the whole of the three-volume first edition himself.[7] The Durants needed two people and roughly four decades for the eleven volumes and some two million words of The Story of Civilization.[8]
Different centuries, different subjects, the same asymptote: a few million words, a few decades, one worldview holding it together. The constraint was mechanical. A person can only read, digest, and transcribe so fast, and a lifetime is a lifetime.
Compilation was always the craft¶
It is worth being honest about what these people were doing, because it flatters them less than the word "author" suggests and illuminates more. They were not, for the most part, inventing. They were compiling — synthesizing thousands of prior sources, quoting, condensing, arranging. Pliny's own preface claims twenty thousand facts drawn from some two thousand volumes.[1] Vincent's Speculum is a vast tapestry of other men's sentences. Britannica's first edition is a digest of the era's available treatises. The encyclopedia has always been, at bottom, the world's existing knowledge passed through one filter.
Which means the scarce input was never the facts. Facts were lying around in libraries. The scarce input was the filter — the unifying judgment that decided what to include, what a thing was, where one topic ended and the next began, and which of two nearly-identical accounts was the true one. That judgment is what made a hundred thousand disparate entries read as a single work rather than a heap. The transcription was the tax you paid to express it. The judgment was the thing itself.
When words become free¶
The mechanical constraint is now gone. A capable language model emits fluent, organized, superficially authoritative prose at a marginal cost approaching zero. The lifetime of reading-and-setting-down that capped Vincent at four and a half million words is simply not the bottleneck anymore. On the raw axis of volume, the compiler's ceiling has lifted by something like an order of magnitude, and the throughput figure is almost comic: where the most prolific compiler in the Western canon sustained perhaps a hundred and fifty thousand words a year, the same lone-curator role can now move at a scale that used to require a funded institute and a room full of contributors.
And here is the turn the essay exists to make: that is the boring half. When words become free, words become exhaust. Volume stops being evidence of anything at all — the open internet is already drowning in machine-generated text precisely because generating it costs nothing. A twenty- or thirty-million-word reference work is, by itself, no longer an achievement; it is a byproduct. If the lifting of the ceiling meant only that one person can now emit more words, it would be a footnote about productivity, and a slightly depressing one.
The scarcity relocates¶
The interesting claim is that removing the mechanical tax does not make the compiler obsolete. It relocates the scarcity onto exactly the faculty that was always the real work, and makes that faculty matter more, not less.
When generation is free, the value of a reference work collapses entirely onto its coherence: whether its categories carve the world at defensible joints, whether a term means the same thing on page nine thousand as on page nine, whether two entries that describe the same underlying pattern are recognized as the same and two that merely rhyme are kept apart, whether the whole structure stays free of contradiction and circularity as it grows. None of that is something a generator supplies. A model will happily produce ten thousand fluent, locally-plausible, globally-incoherent pages; keeping them coherent — holding the single sensibility across the whole — is human judgment doing the one job it was ever really doing. The scribe's burden is lifted. The editor-in-chief's burden is now the entire job.
So the "one-author encyclopedia," far from being a quaint casualty of the AI era, becomes newly possible at a scale its own two-thousand-year history could never reach — but only for an author willing to do the part that was always hard. The tool multiplies output; it does not supply taste, and it actively tempts you to accept fluency where you owe structure. The discipline that used to be spent transcribing must now be spent entirely on judgment, or the result is merely large.
An open experiment¶
This Encyclopedia is, among other things, a live test of the proposition. It has grown past the historical ceiling on the cheap axis — the word count now sits an order of magnitude beyond Vincent of Beauvais — and the essay's whole argument is that this fact should be read as a statement about the new cost of words, not as a claim of merit. The number is the exhaust. Whether the thing is a cathedral or a very large pile is a question that volume is constitutionally unable to answer; only coherence can, and coherence is judged slowly, entry by entry, boundary by boundary, by whether the structure holds when someone leans on it. That verdict is not in, and it should not be rushed.
What the lifted ceiling really offers is not a bigger monument but a harder and more honest job than the old compilers had. They could hide a great deal of missing judgment behind the sheer heroism of the transcription; the years of labor were themselves a kind of proof. That alibi is gone. When the words are free, there is nothing left to be impressed by except whether the thing is true to its own structure — and that has always been, and now nakedly is, the only part that was ever worth doing.
References¶
[1] Pliny the Elder. Natural History (Naturalis Historia). Rome, AD 77 (Books I–X; the remainder published posthumously). Thirty-seven books, roughly 400,000 words; in the preface (Book I) Pliny claims to have gathered some 20,000 noteworthy facts from around 2,000 volumes by 100 select authors. The canonical example of the encyclopedia as one compiler's synthesis of prior sources, and the source for the ~400,000-word and "two thousand volumes" figures in the text. ↩
[2] Pliny the Younger. Letters (Epistulae), Book III, Letter 5. c. AD 100–109. The nephew's account of his uncle's working method — a secretary reading aloud and taking dictation through nearly every waking hour, including at meals and after the bath — supporting the description of the sheer labor the compilation demanded. ↩
[3] Isidore of Seville. Etymologiae. Compiled c. 615 – early 630s; issued in final form shortly after Isidore's death in 636. A single-author encyclopedia of late-antique knowledge that remained a "basic book" of Western learning for the better part of a millennium; its survival in more than a thousand manuscripts is second only to the Bible. Supports the claim about Isidore's long medieval influence. ↩
[4] Vincent of Beauvais. Speculum Maius. Compiled c. 1235–1264. Eighty books and 9,885 chapters; the complete printed edition — including the Speculum Morale, added in the 14th century by later hands — extends to about 4.5 million words, the largest reference work produced in the West before roughly 1600. Supports the ~4.5-million-word high-water mark and the ~29-year span. ↩
[5] Johnson, Samuel. A Dictionary of the English Language. London, 1755. 42,773 headwords illustrated by around 114,000 quotations, prepared over roughly nine years and very nearly single-handed — the first English dictionary to substantiate its definitions with quotations. Supports the headword, quotation, and timespan figures. ↩
[6] Webster, Noah. An American Dictionary of the English Language (1828). About 70,000 entries compiled over some 28 years, during which Webster studied 26 languages to trace etymologies. Supports the entry count, timespan, and languages-learned figures. ↩
[7] Encyclopædia Britannica, first edition. Edinburgh, 1768–1771. Three volumes, roughly 2,400 pages, issued in 100 weekly parts; almost the entire text was compiled by a single editor, William Smellie, with only minor outside contributions. Supports the claim that Britannica began as a near-solo effort. ↩
[8] Durant, Will, and Ariel Durant. The Story of Civilization. New York: Simon & Schuster, 1935–1975. Eleven volumes, nearly 10,000 pages and about 2 million words, written over roughly four decades by two authors. Supports the ~2-million-word and multi-decade figures and the two-author qualification in the text. ↩