Gemini 3.0 Deciphered the Mystery of a Nuremberg Chronicle Leaf's
Key topics
The mystery surrounding a 500-year-old Nuremberg Chronicle leaf's marginalia has been reignited by Gemini 3.0's analysis, sparking a lively debate about the AI's capabilities and the validity of its findings. While some commenters, like game_the0ry, are thrilled to see AI being used in this way, others, such as KaiserPro and BigTTYGothGF, express skepticism, questioning the lack of verification and the assertion that the marginalia was a mystery to begin with. As the discussion unfolds, it becomes clear that the absence of expert citations and peer review is a major point of contention, with some, like why-o-why, dismissing the findings as "fancy slop." The conversation takes an interesting turn when brokensegue and ajross chime in, highlighting the complexities of deciphering ancient texts and the potential differences between this case and deliberately obfuscated manuscripts like the Voynich manuscript.
Snapshot generated from the HN discussion
Discussion Activity
Very active discussionFirst comment
4h
Peak period
28
0-12h
Avg / period
7.8
Based on 31 loaded comments
Key moments
- 01Story posted
Jan 1, 2026 at 1:08 PM EST
7 days ago
Step 01 - 02First comment
Jan 1, 2026 at 4:55 PM EST
4h after posting
Step 02 - 03Peak activity
28 comments in 0-12h
Hottest window of the conversation
Step 03 - 04Latest activity
Jan 7, 2026 at 9:54 AM EST
1d ago
Step 04
Generating AI Summary...
Analyzing up to 500 comments to identify key contributors and discussion patterns
Want the full context?
Jump to the original sources
Read the primary article or dive into the live Hacker News thread when you're ready.
There isn't verification, and its based on the assertion that this marginalia is a mystery. None of which appears to be backed up.
It then doesn't actually do any analysis of the output, any verification, just pastes the dumps at the end, with no attempt to make it readable.
I find this a little hard to believe.
But what immediately comes to mind from reading the title are all the "AI solutions" for the as-of-yet undecoded voynich manuscript that are posted with surprising (and increasing) frequency to at least one forum. They're all incompatible and fall apart on closer inspection.
A collection of them can be found at https://www.voynich.ninja/forum-59.html .
The notes in the linked article are presumptively-legible notes made in good faith, just not with enough detail for someone-who-is-not-the-author to understand . AI training sets are much broader than mere human intuition now.
1. Let's start with where the post was published. Check what kind of content this blog publishes - huge volumes of random low-effort AI-boosting posts with AI-generated images. This isn't a blog about history or linguistics.
2. The author is anonymous.
3. The contents of the post itself: it's just raw AI output. There's no expert commentary.
This isn't to say that LLM aren't useful for science; on the contrary. See for example Terrence's Tao blog. Notice how different his work is from whatever this post is.
The most real benefit of HN vs Reddit is commenters who are actually knowledgeable in that field, who leave a comment or vote up an actually useful comment.
My colleagues do this as well with AI and it fucks me right off.
They just present the raw output, in its long form an expect everyone to follow the flow. Context is everything, damn it.
Looking into it further there isn't really a mystery as to what they are, or at least none that I could find suggesting that its unknown. Especially given the context of the page.
Its great that gemini can do this, its a shame that lots of the ancillary "analysis" about the writing doesn't appear to be correct (humanist minscule I would suggest is too new, too heathen and too Italian for a german manuscript of the time https://medievalwritings.atillo.com.au/whyread/paleographysu...)