60 years, seven planned volumes: Knuth is still writing The Art of Computer Programming
For decades Knuth has mailed hexadecimal reward checks, $2.56 apiece, 'one hexadecimal dollar', to anyone who spots an error in his books. One computer scientist in the thread called getting one a career highlight.
Volume 4B ran 714 pages when it landed in 2023; a fascicle on constraint satisfaction followed in 2025. The page still lists Volumes 5, 6 and 7, compiler techniques last, as work Knuth intends to write.
Then a reader surfaced Knuth's own recent preprint. An AI had solved a conjecture he'd left open, and he wrote it was time to 'revise my opinions about generative AI', calling it a joy rather than a threat.
The unfinished monument that keeps announcing more volumes
Knuth's page for *The Art of Computer Programming* reads like a progress log kept over half a century. Volumes 1 through 3 were finished in the 1990s. Volume 4A arrived in 2011, Volume 4B in 2023 at 714 pages, and a 281-page fascicle on constraint satisfaction in 2025. Still ahead, per the page: the rest of Volume 4, then Volume 5 on parsing, then a plan to loop back and revise 1–3, and only after that Volumes 6 (context-free languages) and 7 (compiler techniques): "but only if the things I want to say about those topics are still relevant and still haven't been said."
The book runs its own economy. Knuth pays $2.56 (one hexadecimal dollar, 0x$1.00) for each error a reader reports, and $0.32 for a worthwhile suggestion. In 1999 American Scientist named the series one of the twelve best physical-science monographs of the century, alongside Einstein, Feynman and Mandelbrot. The full set of authorized eBooks runs about $340, and Knuth asks readers not to report eBook typos unless the same mistake appears in print.
The checks, the humor, and the assembly nobody agrees on
In the thread, the reward checks came up first, and not as a gimmick.
"Receiving one of Don's cheques ('Bank of San Serif') a few months after pointing out an error has been many a computer scientist's career highlight!" — an AI research professor who wrote his first compiler in high school
The reputation is forbidding; the reading experience, several said, is not.
"even if you just dip into it like a coffee table book it's a wonderful read that breaks up tough sections with humor" — a reader who went through the first two and a half volumes cover to cover
The code in TAOCP is written in MIX, and its successor MMIX, Knuth's own idealized assembly language. That choice still splits readers.
"The valuable prose aside, I never liked that code examples were in a pseudo assembly language." — a longtime reader in the thread
Knuth's own note on the machine that beat him to a proof
The sharpest turn came from Knuth himself, surfaced by a commenter tracking his latest preprints. In a note dated this past February, titled "Claude's Cycle," Knuth described an automated system solving a conjecture he had posed:
"It seems that I'll have to revise my opinions about 'generative AI' one of these days. What a joy it is to learn not only that my conjecture has a nice solution but also to celebrate this dramatic advance in automatic deduction and creative problem solving." — a commenter who tracked down Knuth's 2026 preprints
The temperament is the story. A man who budgets decades per volume, and mails a $2.56 check for a single typo, watched a machine crack in one short paper a problem he had left open, and reached for the word "joy." Volume 7 is still decades away by his own schedule. The conjecture did not wait for it.
Get breakdowns like this every day
The top Hacker News stories, unpacked — by email, Telegram, or RSS.
A new runtime for kdb+'s language ships closed-source, benchmark-free, and 'likely AI'
k and q are the terse array languages behind kdb+, the database Wall Street uses for tick data and Kx bills as the fastest and most expensive anywhere. 'l' claims to run existing k/q/qSQL unmodified, with SIMD and compression baked in.