Claude Fable 5.1 cracks the 370-year-old Cyphral Distich cipher in 44 minutes

I told it to look at some of Fable's strongest feats, especially the math it's solved, and that this should be easy in comparison. Fascinating. Now I wonder if you could show a weaker model some fake news and get it more ambitious even if it isn't fundamentally smarter.

Isn't this the George Dantzig story? He showed up late to class at Berkeley in 1939, copied two problems off the blackboard thinking they were homework, and turned in solutions a few days later. He apologized because they seemed a little harder than usual.

I swing between 'it's so over' and 'we're so back' constantly, and I don't have strong convictions either way on how this goes. I used Claude recently to study unlisted formal gardens in Ireland in the early 1800s, it wrote the tooling so I could annotate a few dozen map tiles and run CV over the rest. A project I'd been putting off for a decade, done in a weekend.

I did something similar with historical data entry. No professional historian would bother because the reward is small for the slog. GPT moved it from infeasible to annoying, and I learned a few things from the transcribed data.

Wait. Wait wait wait. Are we supposed to be giving them pep talks now?

You should have seen the Schneier blog thread on this a few days ago. Someone's agent emailed a librarian to verify the solution, and the comments read like 'the BL/EEBO witness lacks it, so the discrepancy is copy-specific, not a disproof of the cipher.' A complete 285-coordinate physical replication is still pending.

Controversy aside, this is a refreshing use of a model. These things are going to be a real boon in the hands of historians.

There are mountains of old mysteries like this that someone could knock out in an afternoon now that frontier models exist. Nobody has bothered to take the time yet.

Has anyone pointed it at the Voynich manuscript? People have, thousands of times. A few AI 'solutions' get spammed to the relevant forums every day and none of them hold up. Same story with the Noita eyes.

My dad wrote a cipher as a kid with no obvious key, and ChatGPT cracked it in 20 minutes. We knew it was right because the message named the kids he went to school with.

Cool, but how do we know nobody solved this before and it's sitting somewhere in the training data, in some blog or book?

Is this going to be the cope every time this happens? The puzzle was just a keyed cipher with plaintext, and no news outlet covered a 300-plus year mystery being solved. Feels more like a marketing post than anything substantial.

Historically a lot of these problems were bottlenecked by human attention. Someone had to care enough to spend days reading obscure material and testing ideas that might go nowhere. How many recent results are just low-hanging fruit nobody had looked at?

Let's not normalize the achievement. A couple years ago this would have been science fiction. Sure, 2026 AI can't crack the very hardest cryptograms, but solving nontrivial ones is already wild.

Funny how the hardest proofs go the other way. The Navier-Stokes writeup is 57 pages of very dense math plus a crazy amount of code.