How 2,000 Servers Broke the Enigma Code in 13 Minutes
blog.digitalocean.com
blog.digitalocean.com
Please use the original title: "How 2,000 Droplets Broke the Enigma Code in 13 Minutes".
"Modern AI" has very little to do with the performance here. As the article states, the team worked 4 weeks and had 2,000 machines to run their code. Even brute-forcing would give you a performance close to that.
Quoting from the article:
> So for the AI to shine, we actually use 2,000 minions that do the tedious work. Everybody praises AI, but it's actually the minions that do the 99% of work. Life, right?
A decrypted enigma message looks like this:
BOOTKLARXBEIJSCHNOORBETWAZWOSIBENXNOVXSECHSNULCBMXPROVIANTBISZWONULXDEZXBENOETIGEGLMESERYNOCHVIEFKLHRXSTEHEMARQUBRUNOBRUNFZWOFUHFXLAGWWIEJKCHAEFERJXNNTWWWFUNFYEINSFUNFMBSTEIGENDYGUTESIWXDVVVJRASCH
Plaintext with word divisions, and with garbled letters corrected:
BOOT KLAR X BEI J SCHNOOR J ETWA ZWO SIBEN X NOV X SECHS NUL CBM X PROVIANT BIS ZWO NUL X DEZ X BENOETIGE GLAESER Y NOCH VIER KLAR X STEHE MARQU BRUNO BRUNO ZWO FUNF X LAGE WIE J SCHAEFER J X NNN WWW FUNF Y EINS FUNF MB STEIGEND Y GUTE SICHT VVV J RASCH
Formatted German plaintext:
Boot klar. Bei “Schnoor” etwa 27. Nov. 60 cbm. Proviant bis 20 Dez. Benötige Gläser, noch 4 klar. Stehe Marqu. BB 25. Lage wie “Schaefer”. NW 5, 15 mb steigend, gute Sicht. Von “Rasch”.
English translation (words in square brackets added to make the meaning more clear):
Boat clear. [Will rendezvous] at Schnoor about 27 November. 60 cubic meters [fuel oil remaining], provisions [sufficient] until 20 December. Need binoculars; [only] four now serviceable. Am at naval square BB 25. Situation like Schaefer’s. [Wind] northwest [force] 5, [atmospheric pressure] 15 millibars rising, good visibility. From Rasch.
> the sum total of all the computing power produced by all of the Bombes during the entire war is about the same computing power used by one iPhone to play back 10 seconds of YouTube
https://www.quora.com/What-was-the-computing-capacity-of-Ala...
What, the proprietary cloud technologies are now re-writing the English vocabulary for just an itsy bit of extra vendor lock-in?
I find this double plus hydrating.
Also it probably would be trivial to write a non-ai that detects valid german (e.g. is every word german?). False positives would be easy to filter out.
By the sound of this article, it makes me suspect DigitalOcean doesn’t really do any machine learning (here they paid for a marketing piece 7 months later?).
Not that they need to: they are a great cloud provider. I wonder why they need to do this for marketing. If the goal is to say, look how we used 1-click for this PR fluff, will that really be attractive to the sort of enterprise ML customers who would possibly care? And why 7 months later?
[1] https://enigma.hoerenberg.com/index.php?cat=Unbroken [2] http://www.bytereef.org/m4_project.html
"What would Alan Turing be able to do nowadays if he had the current computing power and all the development around AI?"
My own impression, which has only grown stronger with the passage of years, is that if you did the totally necessary CRUD stuff as simply as possible, that would open up opportunities to tech-enable businesses in much more interesting and profitable ways. But, of course, that would involve delaying gratification, and often doesn't play to the prevailing politics.
Interesting software jobs do exist but, boy, do you have to hunt for them.
What about getting you paid on time. Or allowing a person to make a contactless payment in another country.
The world is awash with interesting applications of intention technology.
IBM 601 (1936) - mechanical multiplication, punched card I/O
IBM 602 (1946) - mechanical multiply and divide
IBM 603 (1946) - first commercial electronic arithmetic, a vacuum tube-based 602
IBM 604 (1948) - Next-gen 603, a limited computer programmed with a plugboard board
IBM 605 (1949) - a bigger 604
IBM CPC (1949) - An IBM 605 with peripherals - card reader, card punch, memory units, printer.
IBM 607 (1953) - The CPC in more compact packaging with more memory.
IBM 650 (1953) - First real computer in this series. First mass-produced computer. Drum main memory. First disks. Knuth learned on one of those.
IBM 1401 (1957) - Transistorized, core memory computer. Not much memory, but good tape drives. Mass produced.
This was cost-effective computing for business. There was other stuff going on in parallel, but most of it was military-funded. These machines ran business and had to pay. The big problem was memory cost. Everybody knew more memory would make more things possible, but memory was a million dollars a megabyte well into the 1960s.