Fable 5.1 Solves the Cyphral Distich, a 370-year-old cipher
Anthropic's Fable 5.1 cracked a 370-year-old cipher by uncovering its subtle internal key, a puzzle long overlooked by humans. This feat has ignited Hacker News, sparking fervent debate over whether AI is truly intelligent or just a superior pattern-matcher, and what this means for tackling other historical mysteries.
The Lowdown
A recent blog post from Vals AI details how their Fable 5.1 model successfully deciphered the "Cyphral Distich," a 370-year-old cryptogram left by Sir Thomas Urquhart that had stumped human cryptographers for centuries. The solution highlights AI's novel approach to historical problems.
- The "Cyphral Distich" consisted of two lines of 32 numbers each, found in Urquhart's 1653 work, Logopandecteision. It was considered one of Klaus Schmeh's Top 50 unsolved encrypted messages.
- Previous human attempts focused on external keys (cipher alphabets), but Fable 5.1 realized the key was internal: each number corresponded to a word index within Urquhart's accompanying "32 Proquiritations."
- The decoded message revealed a royalist prayer: "O GOD UPHOLD KING CHARLS THE SECOND AND MAKE HIM THE SUPREME RULER OF THIS LAND," consistent with Urquhart's known political leanings.
- Using a similar context-indexing method, Fable 5.1 also largely deciphered a larger, related cryptogram, the "Cyphral Octastich," found in Urquhart's The Jewel.
- The author notes that Fable 5.1's success wasn't due to extraordinary cryptanalysis but rather its persistence in exploring possibilities and noticing a "simple in hindsight" clue that humans had missed, suggesting AI's ability to tackle problems bottlenecked by human attention.
This demonstration suggests a new frontier for AI applications, potentially revolutionizing historical research by sifting through obscure texts and connecting previously overlooked contextual clues to solve long-standing puzzles.
The Gossip
Intelligence or Iteration?
Commenters debated whether Fable 5.1's success demonstrated true intelligence or merely a highly efficient, brute-force-like search through possibilities. While some saw it as groundbreaking problem-solving, others suggested the AI was simply persistent in trying numerous approaches until a 'simple in hindsight' solution emerged, without necessarily 'understanding' in a human sense. The discussion often revolved around defining what constitutes 'intelligence' in the context of LLMs.
The 'Low-Hanging Fruit' Fallacy
A prevalent theme questioned if these ciphers were genuinely 'hard' or simply neglected by human researchers who lacked the time, resources, or motivation to meticulously explore every contextual clue. Critics suggested AI excels at finding 'low-hanging fruit' – problems that are solvable with exhaustive search but were deemed too tedious or unimportant for significant human effort, rather than cracking truly complex, intentionally obscured ciphers.
Elicitation and AI Encouragement
The author's admission of giving the AI a 'pep talk' and telling it to 'think creatively' sparked curiosity and humor. Users discussed the potential impact of such prompting, questioning whether it genuinely influences model performance by shaping its 'metaknowledge' or ambition, or if it's merely a psychological projection onto the AI. Some shared anecdotes of similar experiences with other models.
Ciphering Through the Canon
Inspired by Fable 5.1's achievement, many commenters eagerly suggested other famous unsolved ciphers for AI to tackle. Popular requests included the Voynich Manuscript, the Zodiac Killer's ciphers, and the final, elusive part of Kryptos (K4). This highlights a collective hope that AI could unlock long-standing historical and criminal mysteries that have baffled humans.
AI Hype and Human Humility
The discussion included a broad spectrum of views on the current state of AI. Some expressed profound amazement at the progress, describing a fluctuating 'vibe-o-meter' between utopian optimism and dystopian dread. Others maintained a more cynical or critical perspective, viewing AI as glorified autocorrect or noting its occasional failures in 'trivial' tasks despite solving complex ones, cautioning against excessive hype and emphasizing the distinction between human and artificial cognition.
Climate Calamity: An Uncanny Tangent
A significant portion of the comments veered into an extensive, albeit tangentially related, debate on climate change. This was largely sparked by a user expressing less worry about AI compared to climate change. Discussions encompassed AI's energy consumption and potential to exacerbate global warming, arguments about the effectiveness of existing green technologies, and philosophical musings on humanity's capacity to address collective crises versus its individualistic tendencies. The thread showcased HN's tendency for wide-ranging, sometimes off-topic, but passionate debate.