Fable 5.1 Solves the Cyphral Distich, a 370-year-old cipher
summary
Claude Fable 5.1 successfully deciphered Sir Thomas Urquhart’s Cyphral Distich and Cyphral Octastich by identifying that the keys were embedded within the books themselves rather than being external cipher alphabets. The AI's success highlights a shift where the bottleneck for solving historical mysteries is no longer human attention or tedious archival research, but rather the ability to process and connect obscure clues. This suggests that AI can tackle forgotten conjectures and archival puzzles through persistence and pattern recognition.
key points
- The Cyphral Distich was solved by realizing the numbers in the cipher corresponded to word indices in the author's 32 Proquiritations, with the first letter of each word forming a royalist prayer for Charles II.
- The AI applied the same logic to the Cyphral Octastich, using page numbers as indices to extract first letters of words, effectively solving nearly the entire larger cryptogram.
- The solve was not achieved through complex cryptanalysis, but by noticing simple, context-based clues that human researchers had overlooked for centuries.
who should read it
AI researchers, historians, and cryptography enthusiasts who are interested in how large language models can be applied to solve long-standing historical mysteries.
discussion mixed
consensus
Commenters largely agree that modern LLMs, particularly Claude's Opus and Fable, have developed a frustratingly dense, unnatural, and often ambiguous writing style referred to as 'GPT/Claude speak'. Many users find this linguistic degradation annoying or misleading, though some hypothesize that this verbosity acts as a 'rail' that helps models maintain focus during complex task execution. There is also a shared sentiment that these tools are invaluable for niche hobbyist projects that were previously too tedious for humans to perform.
top insight
A significant observation is that LLMs often lack accurate metaknowledge regarding their own capabilities, meaning they may fail at a task simply because they believe it is impossible. Consequently, users can actually improve performance by providing 'pep talks' or framing problems as solvable, which pushes the model to attempt solutions it would otherwise dismiss due to biases in its training data.
Generated by an LLM from the article and thread. Check the sources before relying on it.